heres a dumb question
do you work for FFDshow?

Oh yes! The CPU numbers in the OSD are a lot better (they were sort of stuck on 100 %, with this new version they move very fast up and down). I can't say much about the picture quality since I'm on a small computer monitor (20") right now, but the "torture pan" with Gandalf is very smooth, and I believe it's as smooth as it can get. I don't think that long pan is 100 % smooth in the source material. This is very, very nice!

do you work for FFDshow?ffdshow, AFAIK, is an open-source project, so he doesn't work for anyone. If anything, the code works for him and anyone else who wants to mess with it. All hail open source!

;)

who made FFDshow then? and who posts the online updates?

Is there a way to force FFDSHOW to only work with a few resolutions or file types?

IE - I only want it to work with DVDs - not process HD content.

Or at the least - can you run 2 instances?

Are you guys uninstalling previous versions before reinstalling Andy's betas, or installing over them?

Hawkson: The person who started the project was a Milan Cutka (http://sourceforge.net/users/milan_cutka/), who states that the project "is based on original DirectShow filter from XviD, which is GPL'ed educational implementation of MPEG4 encoder," and is an extension of the ffmpeg project. The last alpha was date May 23, 2003, so I'm guessing others have taken the project and ran with it like AndyIEG did.

I'm simply installing over the previous instally.

Andy, just as a FYI, the latest SSE2 version works nicely... no problem to report with Theatertek

On my Celeron overclocked to 2.66GHz, I can run denoise3d followed by resize at 1440x960 on NTSC DVDs using Owen's suggested settings. I'm still a little bit low on power to resize PAL discs to 1440x1152 (resizing to 1024x768 does work well with an average CPU utilization of 80%)... but this may be asking too much

Thanks again

Eiffel

I have installed ZP3.31+FFDshow(20030523)+Sonic filters. I am having problems with FFDshow. ZP works fine until I select FFDshow. I get multiple pictures simultaneously.
Has any one experienced that, and what is the remedy?

radek

Andy, did you slip in an SSE2-optimized DScaler Sharpness filter into the package as well, or is that from something else? If so, you're a tricky devil, and not to be trusted with our valuable HTPCs' tender processing power ;) .

I noticed something interesting tonight. I normally use Sonic for audio decoder. Using Sonic for Video and Audio I was able to Denoise 3d, Resize to 1776X1000, and the playback was smooth with CPU hovering between 40 and 67% on the FFDSHOW OSD. When I registered the AC#X filter in place of Sonic, it choked my system up and CPU was at 100% and playback was very choppy???

Originally posted by JBlacklow
Andy, did you slip in an SSE2-optimized DScaler Sharpness filter into the package as well, or is that from something else? If so, you're a tricky devil, and not to be trusted with our valuable HTPCs' tender processing power ;) .

i did "slip" this in since dscaler is also GPL'ed so why not add the dll only, since most only use the sharpen filter. So u dont need to install dscaler only for the sharpen filter. While i was building the .dll i was thinking why not rewrite the mmx code fast to a sse2 version :)


"who made FFDshow then? and who posts the online updates?"
Milan is still making all the main work, but he rather change the code or add stuff than compiling the final package u use. Athos is doing this and does a nice job releasing the package.

Milan is adding stuff to the cvs project tree and he is still working on the code. He dont update the "official" homepage anymore but Athos and i get the code fresh out of the cvs tree and build the final package and filter based on this. We than also started to add little fixes wich dont made it into the cvs and latly i started this SSE2 optimisation stuff myself.
Milan still does the main job on ffdshow :)

Think of it like xvid Milan=official ffdshow and Athos and me are like Nic's and Koepi's xvid release sites.

Originally posted by Eiffel

On my Celeron overclocked to 2.66GHz, I can run denoise3d followed by resize at 1440x960 on NTSC DVDs using Owen's suggested settings. I'm still a little bit low on power to resize PAL discs to 1440x1152 (resizing to 1024x768 does work well with an average CPU utilization of 80%)... but this may be asking too much

Thanks again

Eiffel

Don't you think if you can run that on a Celeron I can run that on an AMD Barton 2500+? I get choppy playback with the CPU at 100% using the same settings.

i can't run that on my 2400+ clocked to 2700 (2166mhz)

Originally posted by Spoonfed
i can't run that on my 2400+ clocked to 2700 (2166mhz)


So what are you using? Gradual?

Yep gradual denoise and resize (lancoz) to 1024 x 768.
Also Unsharp after the denoise

About 75-80 CPU (with spikes). Can get a little more resize but need some spare for muliti dtv card captures etc (while FFDShowing recordings).


A little off topic but my Hightech 9600 Pro is dying (give BOSD and windows says driver stuck in infinite loop, windows then says usually indicates hardware failure (use have upgraded graphics and MB drivers)

Anyhooo its under warrenty, BUT i need PC while its in for RA. Willl say a 9200SE (Gexcube) do the job for the time being (HD/SD/FFDShow) etc? I don't use PC for gaming.

Theres heaps of CPU talk but im not certain if such a graphics card would be enough?

AndyIEG, thanks for the great work.

I have:

P4 3.06GHz overclocked to 3.4 GHz
533MHz FSB
2 x 256 MB PC2100 DDR (266MHz)
Hyperthreading turned off

For the first time I'm able to run resize at double DVD (1440x960) without shuddering. I'm running TheaterTek with the following filters with smooth playback (in order shown):

Picture Adjust - reducing saturation
Denoise - denoise3d (Owen's settings)
Resize/sharpen - 1440x960, Lanzcos (Owen's settings)

When I had the OSD active I was showing frequent spikes to 100% CPU @ ~22fps with some shuddering. When the OSD was deactivated, the reduced load gives a pretty smooth playback, and my LCD display tells me I'm between 75% and 85% CPU.

I was pleasantly surprised that resizing to 1440x960 gives a noticeably better picture than the 1280x720 that I was limited to before, despite the fact that my 42" plasma display is 852x480.

The only problem that I'm having is TheaterTek crashes frequently when exiting the program. Not every time, but frequently. Never had that problem before.

Thanks again,
Phil

I have had a few crashes in ZoomPlayer while starting playback, especially if I reboot the computer and then start playback. When I close Zoom after the crash and start playback again, it usually works. I guess I'm officially at the bleeding edge now, since my HTPC is not allowed to crash at all normally (but since my wife and kids are away for a few weeks I can experiment...).

Windows shows 86-89 in CPU use, while the ffdshow OSD of course is too fast to follow. Something weird in the OSD, though: It jumps between 31 and 22, it seems (again fast numbers, hard to follow). But I'm playing PAL which should be 25 FPS. So what's up with that? It does not change if I remove all filters, the numbers jump as fast then.

@pmd918
The only problem that I'm having is TheaterTek crashes frequently when exiting the program. Not every time, but frequently. Never had that problem before.
@Mastiff

So those crashes apeer if u close the player?

Do these crashes only appeer with my latest version's? Can u try the normal version too if u used the P4?

PS: oki found a little bug in the P4/A64 version if the X resolution is not divisible through 16...(fixing it soon)

The crashes happen when I start playing, but they're not dependable. Sometimes it crashes, sometimes it doesen't. Also I can't get ffdshow working with Sonic (I installed that today just as a test reference). I will do a backup of my system (PQ DriveImage) and then try the normal version.

OK, it's dependable... Whenever I reboot the computer and start playing a DVD ZoomPlayer will crash the first time. A weird crash, since it doesn't dissappear at once when I click the "close program" button on the crash dialog box, it takes up to 20 seconds. The second time I start playing it works. And now I can tell you that this happens only with the P4 version, not the regular version. I'm using WinDVD video and I have experimented with a few audio filters, but that doesn't matter.

In the OSD there's not much difference between the regular and the P4 version, but I can see that playback is more jerky with the regular version.

Some time ago somebody mentioned a setting that could show half the picture untreated. Is this in there somewhere now? I can't find it...

mhh oki i think i found the bug for the zp crashes. The function is in every filter tab (top righ corner "process only right half")

Good! And thanks...I need to get a refund on that LASIK operation five years ago... ;) I was only looking in the Resize and didn't notice it since it was greyed out... I wish it was possible there as well, but I can see why it isn't.

Just an FYI, I haven't experienced any crashes with the current latest SSE2 build and the latest version of TheaterTek (I do have some problems with girder when CPU utilization goes to 100% for too long, but I don't think this is an FFdshow issue per se).

Eiffel

i had some crashes using the latest reclock filter as sound render maybe this is the problem for u Mastiff? (try deinstall reclock and test) thx

No, I don't use it. I have it installed, though.

Andy,
I'd like to test the sse2 version but i only see one .exe to the link you provided.
Is that the sse2 optimized version?

Originally posted by vkon
Andy,
I'd like to test the sse2 version but i only see one .exe to the link you provided.
Is that the sse2 optimized version?

doing some bugfixes on the SSE2 version atm

Just when you thought you had enough CPU power for FFDShow, I have upped the anti to a new level, just to keep your CPU’s maxed out. :D

Thanks to Andy’s great work optimizing my favorite FFDShow filter (Denose3d), I found myself with CPU power to burn so I decided to go crazy with resize.

I experimented with Avisynth Lanczos and Lanczos4 resize but could only get 1440x1152 to run smoothly which limited effectiveness for my 1776x1000 display resolution. Avisynth Lanczos is a little sharper then FFDSHow Lanczos at 0.0, but I found that FFDShows Lanczos filter at 1920x1152 works better for my PAL DVD’s with less color artifacts on fine textures like clothing and hair.
The Avisyth Lanczos filters are too CPU hungry for 1920x1152 operation.

Anyway since Andy’s fine tuning of Denise3d (thanks again Andy) left me with only 60% CPU usage with Denoise3d followed by Lanczos resize to 1920x1152 with Elecard decoder (2510 build) and Overlay output, I decided to go higher to 1920 x 1728 or 2.66x DVD width by 3x PAL DVD height.

No that is NOT a miss print; I am using Denoise3d followed by resize, (Lanczos) to 1920x1728 with absolute smoothness.
All I can say is WOW, almost all traces of the digital video look have been removed, leaving a VERY smooth and artifact free image that looks very much more like film.
Fine detail appears better and the color artifacts I was getting on fine textures like clothing and hair are gone. Even background objects look better defined.

Double DVD resolution has become the defacto standard for resizing, but now I am challenging that assumption. Triple DVD resolution is definitely better on my display.

FFDShow is limited to 2000x2000 for some strange reason so I cannot use 2160x1728 for PAL although I believe my system can handle it.

Andy, have you any idea why there is a 2000x2000 limit on resize in FFDShow?

So to wind up, if you have a fast system, try Denoise3d L0.5, C1.0, T5.0 HQ followed by Resize (Lanczos) to 1920x1728 for PAL or 1920x1440 for NTSC.
Resize sharpen settings (if used) are up to you.

Have fun people, it keeps getting better.
I really believe that those expensive stand alone scalers have met there match now.

Buy the way Andy, your 20040607a build is about 5-10% slower on my P4 system then the 20040529a build.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

I haven't noticed any performance improvements from this new 0607 build over the 0601c build, but I'm doing my testing on my main P4 3.3GHz system rather than my much slower HTPC, and this is on xvid encoded videos.

@Owen and Goi

are u using the "ffdshow-20040607a_P4_AMD64.exe" or "ffdshow-20040607a.exe" version?

I removed the P4/AMD64 version for now since i want to fix a serious bug. The "ffdshow-20040607a.exe" dont have any SSE2 code, its a normal i686 MMX2 version.

Andy,

My mistake, I am using "ffdshow-20040607a.exe".
It does seem a little slower then the older build though.

Have you any Idea why there is a 2000x2000 limit on resize?

Regards,

Owen

maybe just a simple max in the calculation check or the array's holding all the informations are hardcoded for just 2000 values max... i will check this later maybe


PLS if u find bugs or stuff related to the P4/AMD64 version send me private messages. Lets keep the crash talk small in the thread

thx

Owen,

The more the better is true with resize. I can never get enouth. However, I realy can not switch to overlay (which would give me power to spare) as my Nvidia overlay is clearly inferior to color rendering with VMR7 (while colors can appear more saturated, they are not rendered with such precision and gradations of colors). Overlay also introduces a bit more noise and I can notice individual pixel structure from 1.5 diagonal of my 61" screen. I know you did it before, but it would be great if you could try VMR9 again and report back how ATI overlay compares to VMR7 or 9 in terms of color rendering and noise (preferably without Denoise filters). I would really appreciate your "last word" on overlay vs. VMR renderers.

Also, as I reported in other thread, I find AA and AF to improve overall image quality in VMR7 setup with 1920x1080 FFDShow Lanczos resize. It would be great if you could try it out and let us know what you think. My Nvidia Quadro really does a great job with AA in particular. It is difficult to get information of how high does AA resampling go (it is "dependent on input resolution") but I am at least sure that 4xAA takes image to 3840x2160 (as its max output resolution is 3840x2400) and samples colors at that resolution. It would be interesting to hear how is ATI doing in that regard. Even in the new line up (X800 vs. 6800) reviewers give ATI a slight edge in AA sharpness. By the way, my Quadro with 61.12 Beta drivers is using new AA (available in 6800 series of cards) algorithm "rotated grid AA" as opposed to previous geometric 4 x sampling.

My X800XT is due in later this week, so I'll give you feedback on how it looks ;)

I second that request on Overlay quality of ATI and nVidia. There is another thread on the CRT Projector forum about how the nVidia 5950 is a cleaner output signal vs any Radeon card.

Owen. You say 3x Native DVD resolution looks much better but wouldn't one only be able to use that if their display was capable of showing it?
I have a Marquee Ehome CRT Projector 8500 with 8" tubes. I was told by the experts that its a waste of time to view native dvd's anything higher than 1440x960. Because reason A, scaling past a certain point gets you nothing more. Reason B, my projector lines on the screen will start to overlap each other. Maybe that last reason depends on how close the projector is to the screen?

Your projector is going to scale the source back down to it's native, isn't it?

ELectrohome Marquee 8500 crt projector.

- 2500 x 2000 pixel addressability
- HD ready with support for all 18 HDTV formats including 1080p

- Horizontal Frequency range of 15 kHz to 130kHz
- Vertical Frequency range of 38 Hz to 180 Hz
- 100 MHz Bandwidth
- Input level .5 to 5.0 volts p to p, 75 ohms terminated
- RBGHV input module standard
- 1200 Peak White CRT Lumens
- 25,000 to 1 contrast levels with below black levels

- High Definition HD-8 Hybrid F1.1 precision lens assemblies
- 10 line pairs per mm resolution
- Scheimpflug focus balance for accurate corner focus
- Electromagnetic focus CRT's
- Hi-output, air-coupled CRT's
- High 100Mhz bandwidth electronics to maintain razor sharp image quality
- Cast aluminum lens mounting maintains exact lens positioning and dissipates heat to minimize distortion

His is a CRT and thus doesn't have a native resolution. It will display at whatever frequency it is given up to its limits, even past its optimal resolution.

I hope to try this out tonight, I have been using FFDshow for a while now though haven't messed with it much as of late. I have a mobile P4 in my htpc now and it is doing ok but could deffinately use some help with the SSE2 optimizations.

It's still going to scale it back down to the max resolution (or whatever resolution you force the display to).

I just had an Oh Sh%t moment.

I've been wondering for a while why my playback performance using TheaterTek and ffdshow seems to be lower than what others report, and what I'd expect for my equipment.

I just realized that most people seem to be running reClock. A month ago I bought my first Dolby Digital receiver, so up until that point I hadn't been using the AC3 decoding in TT. It now occurs to me that performance dropped considerably when I started to output the AC3 stream to my new receiver.

Are most people using reClock? Will it help my performance significantly? Are there other, better solutions? I don't want to add another software program unless it's absolutely necessary.

If this has been my problem all along, boy do I feel stupid!

Thanks,
Phil

Originally posted by madpoet
It's still going to scale it back down to the max resolution (or whatever resolution you force the display to).

Actually, the video card will downsample the image to your actual output resolution (the projector has nothing to do with it). I believe the point here is to resize as high as you can go using FFDShow and then let the video card's scaler downsample again back to your actual output resolution.

That's why you see so many people with native 1280x720 output commenting that sampling to 1440x960 (even though it gets downsampled again) looks much better.

-Pat

I see that point but thats not anything I need to do with my projector. I use Powerstrip to set my output to any res I need it at.
So if I want my Resize in ffdshow to be set at 1440x960 then I set powerstrip to 1440x960.
If I want 3x native dvd res of 2160x1440 then I would set power strip to 2160x1440.

Question I had stated previously is
A: I was told its not recommended to go beyond 2x native dvd res because it would be a waste and wouldnt see any improvement.
B: My projector at that high a resolution would have overlapping scan lines. Im not sure if that has anything to do with how close my projector is to the screen.

Add: Something about resolving so many scan lines per inch? Is that the 10 pair lines per inch stat in the Marquee discription I posted earlier?

Oh true I wasn't even thinking in terms of video card scaling.

Originally posted by stylinlp
I see that point but thats not anything I need to do with my projector. I use Powerstrip to set my output to any res I need it at.
So if I want my Resize in ffdshow to be set at 1440x960 then I set powerstrip to 1440x960.
If I want 3x native dvd res of 2160x1440 then I would set power strip to 2160x1440.

Question I had stated previously is
A: I was told its not recommended to go beyond 2x native dvd res because it would be a waste and wouldnt see any improvement.
B: My projector at that high a resolution would have overlapping scan lines. Im not sure if that has anything to do with how close my projector is to the screen.

Add: Something about resolving so many scan lines per inch? Is that the 10 pair lines per inch stat in the Marquee discription I posted earlier?

Not sure if we had semi-overlapping posts or I may just not be explaining it well enough. The point is to leave your powerstrip settings at the sweet spot for your projector as far as not overlapping scan lines, etc. 1280x720 and 1440x960 are common depending on how good your focus is and the projector. Completely independent of that they are recommending to scale as large as you can in FFDShow (beyond what powerstrip is set for). Your video card will then downsample the image to the resolution powerstrip is set for and the claim is that is a much better looking image than just directly scaling to the projector's resolution.

Originally posted by pcgeek
Actually, the video card will downsample the image to your actual output resolution (the projector has nothing to do with it). I believe the point here is to resize as high as you can go using FFDShow and then let the video card's scaler downsample again back to your actual output resolution.

That's why you see so many people with native 1280x720 output commenting that sampling to 1440x960 (even though it gets downsampled again) looks much better.

-Pat

Depends on how you do it. You can either have the video card scale it down or the projector.

pcgeek. Thank you for further explanation. But wouldn't it be perfered to have ffdshow resize and powerstrip be all at the same resolution? Even downsampling or downsizing creates artifacting.

BTW I asked on the projector forum what my sweet spot would be for my projector at my screen size and distance. I think those factors influence the sweet spot. 8' wide screen 1.85:1 AR.

Originally posted by madpoet
Depends on how you do it. You can either have the video card scale it down or the projector.

Sorry, yes either can do it. I run a CRT so I'm used to it just being a fairly dumb device.

stylinlp: That has been the general understanding here for a while that is currently being challenged. As is often the case here, it doesn't hurt to just try it yourself and see if you like the results. If not, go back. It doesn't actually have to play smoothly to be able to judge the picture quality to some extent so even if you are borderline and thinking of upgrading you should still be able to evaluate (and maybe wait for AndyIEG to make his way to optimizing the resizers).

-Pat

stylinlp,

I thought the same thing, but I can tell you that in my experience, downsizing is much better.

I have an 852x480 plasma. I originally tried Powerstrip at the native resolution of the display, but found that the PQ was significantly better when I set the Powerstrip resolution to 1280x720 and let the display downsize. At the time, I was running ffdshow resize to match the desktop resolution of 1280x720, so there was only one downsize calculation.

With the latest versions of ffdshow, I now have enough horsepower to run ffdshow at 1440x960. I was not expecting to see any PQ improvements, but as I stated in my post earlier today, the PQ improvement was definitely visible. And that's with 2 downsizing events!

Definitely not intuitive. I would have thought that the artifacts introduced with 2 downsize calculations would produce lower PQ, not better PQ. I don't understand it, all I know is that I see what I see!

Phil

Originally posted by AndyIEG
@Owen and Goi

are u using the "ffdshow-20040607a_P4_AMD64.exe" or "ffdshow-20040607a.exe" version?

I removed the P4/AMD64 version for now since i want to fix a serious bug. The "ffdshow-20040607a.exe" dont have any SSE2 code, its a normal i686 MMX2 version.

I'm using ffdshow-20040607a.exe since that's the only executable I can find on the page.

Originally posted by stylinlp
pcgeek. Thank you for further explanation. But wouldn't it be perfered to have ffdshow resize and powerstrip be all at the same resolution? Even downsampling or downsizing creates artifacting.

BTW I asked on the projector forum what my sweet spot would be for my projector at my screen size and distance. I think those factors influence the sweet spot. 8' wide screen 1.85:1 AR.

You are right that both upsizing and downsizing create artifacts i.e. something that is not in the original signal. The problem with DVD signal is that original does not look good on large screens. Just not enough pixels. For example, CRTs can do 740x480 native DVD resolution and output image without any scaling. However the quality of that image will be low on any screen past let's say 40-50 inches, assuming average 2 x diagonal size viewing distance. Scaling is a necesity with DVDs and large screens. It will create artifacts, but if done right will also increase apparent resolution and detail of DVD image.

The goal (at least IMO) is to actually create a new high resolution image, based on the original low resolution image. It is much more than just "fixing" the original DVD signal. In doing that, downscaling upscaled image in digital domain will help you smooth out various artifacts created through the upscaling process. For example, I scale 740x480 DVD image to 1980x1080 Lanczos and then let my card scale back to 1280x720 (native resolution of my 61" DLP RP TV). I like this better than scaling to 1280x720 with FFDShow and not have my card scale the signal at all. Smoother image with ultimately more real detail, and edge enhancement (original DVD transfer as well as some Lanczos ringing) that is less visible. In DVE resolution pattern you should notice ability to resolve more detail if you scale to 1080p. I'd say there is benefit of going with more than 2xDVD resolution. See my previous post in this thread about AA processing (that is sort of chroma scaling equivalent) for even higher resolutions. The benefit is diminishing as you scale to higher and higher resolutions. Kind of the law of diminishing returns.

Also an important factor is coorelation of output resolution to max resize resolution. (downscaling divider). I noticed that going from 1080p to 720p gives you noticeably better image quality than going from 1008p to 720p, although the difference between 1080p and 1008p is less than 200K pixels. Looks like downscaling with 1.5 divider is preferable. I'd say it would probably be even better with downscaling divider of 2 (2560x1440 FFDShow resize for 720p output), but at this point FFDShow is limited to 2000x2000 resize.

Owen, when did you upgrade to a P4 EE running @ 4.5 mHz? ;) I wouldn't be able to go near that resize, I think. But that's probably since I'm using the dScaler sharpen, as I believe you're not. When I get my head above water here (I'm working 16-18 hour days now that my wife and kids are away) I will do a bit of testing to see if it's best on my BG808 to run with more resize and without dScaler sharpen or with dScaler sharpen and 2 x DVD resolution. And yes, you're driving me mad! Just when I thought I had a HTPC that could take everything your twisted mind and ffdshow could dish out... :D

Actually, Vlad, I would say that the goal is to get the best picture you can get, no matter how many downsizing events or upsizing events or even main events you use on it! :cool:

pmd918, you were on the previous page, so you may have been forgotten... I tested earlier today (I'm in a serious period of testing because I'm going from an AMD XP2600 to a P4 3.0 @ 3.75 (have I bragged about that overclocking before...? ;) )) and reclock did not have any effect on CPU usage for me. That was with WinDVD audio filter. Anyway, TT is good as a plug and play thingie, but maybe going to ZoomPlayer and Elecard or WinDVD would help? I believe Zoom has a rather large advantage for this use since it's so configurable. TT is just...TT. Sonic filters, and that's that. :cool:

An idea, guys: How about making it a part of your signature or description what settings you're running ffdshow on, and what processor speed and finally what kind of display device? I will do that when I get the PC up and running with the projector, not just a 20" postage stamp sized computer monitor.

what cooling are you using to get the 3.0 at 3.75? Thinking about getting the 2.8, but can't decide on which one to get.

Zalman CNPS7000A copper. Heavy as h... Two blowholes in the top of the case (as a norwegian I can take anything a whale can dish out...) with 80 mm Papst fans running at 7V, so they makes very little noise. The PSU has another of those Papst fans, but that's on 12V (I deducted that burning my hands on the colling fins of the PSU was not a good thing...) so that's a bit more noisy.

As for you I'd stay away from Prescott if you're not planning on going water or wapo chill cooled since they run a lot hotter, at least that's what I have heard at the overclocking forums. Buy a 2.8C, that should get you to 3.5 without any special measures. :cool:

Maybe you guys can shed some light on this for me.

I followed the guide on HTPCNews for FFDshow (http://htpcnews.com/main.php?id=ffdshowdvd_1) and after a ton of headaches with crap not working and going through [some of] the posts here, I think i've gotten it to work, but im not sure if my results are par for the course. Im having issues taking screenshots/captures (fraps keeps resizing them) and i dont know an objective way of comparing before after (currently i am waiting for a DVI cable to connect to my TV, so this all has to be done on the monitor). The best i could do was this:

http://plaza.ufl.edu/apeboy/ffdshow.gif

The 'without' shot is 100% (and shot in Cyberlink PowerDVD). The with shot was reduced 50% in photoshop CS bicubic (so it matched the default DVD resolution. Taken in Zoomplayer) Do those differences seem about right?

Using Sonic Cinamaster video decoder, ffdshow, zoom player. My FFDshow settings are: resize 1440/960, Lanczos: Luma Sharpen/Chroma Sharpen 2, Parameter 10. Blue and NR: Gradual Denoise 40. Video Renderer: Overlay mixer.

Originally posted by Mastiff
Zalman CNPS7000A copper. Heavy as h... Two blowholes in the top of the case (as a norwegian I can take anything a whale can dish out...) with 80 mm Papst fans running at 7V, so they makes very little noise. The PSU has another of those Papst fans, but that's on 12V (I deducted that burning my hands on the colling fins of the PSU was not a good thing...) so that's a bit more noisy.

As for you I'd stay away from Prescott if you're not planning on going water or wapo chill cooled since they run a lot hotter, at least that's what I have heard at the overclocking forums. Buy a 2.8C, that should get you to 3.5 without any special measures. :cool:

You're right about the Prescott, they're really hot, and MHz for MHz they're slower than the Northwoods in most things. However, not all 2.8C chips will do 3.5GHz. Mine only does 3.4GHz and requires 1.65V to do it, as well as a huge-ass Swiftech MCX4000 HSF.

I have a Nvidia 5900xt and a Sony GWIII. Which one should I set to do the downscaling, the video card or the monitor? Which one will give me the best pq in downscaling?

Currently I have the GWIII doing the downscaling and it looks good, but haven't tried it with the video card yet.

Heh-heh! You might have read this one before, but there's only one way to find out: Try it yourself. It all depends on the quality of the scaler in the video card compared to the Sony's scaler, and my guess is that not many others here has exactly the same combination, so you'll have to do the legwork yourself.

I have no idea what you're complaining about (if that was a complaint), you big ape! (Sorry, I'm drunk... :rolleyes: ) You have gotten a lot out of it. That movie is so incredibly soft that I went totally ape (sorry again) the first time I started to see it. I have NEVER had dScaler sharpen that high before. Without it was a totally soft blurfest!

If what I'm seeing on those images is what you're seeing on your display device, it's worth around 2000 - 3000 dollars in digital projectors. That's how much you would have to pay to get a similar increase in image quality the last time I looked at DLP's. So that would be worth a lot of "craptime" and headaches, believe me! :cool:

Hello, I'm new to ffdshow; I read some guides and many posts here, but I can't find any hint on which should be the best average settings for DivX/XviD playback on a digital screen? (LCD monitor or plasma TV)

No one ever talks about postprocessing settings, for example; is that better than denoise, for compressed files in which there are many blocks?

Should I use Dscaler with DivX? And...by the way, where do I find the Dscaler Sharpen plugin?

Thank you for your attention.

The best place for DIVX/XVID is in your recycle basket... Sorry if I seem harsh, but in this thread we strive to make the not yet good enough DVD images better, which is why there is nothing about even worse formats. Some may be able to answer those questions here, but I doubt it. Maybe you could try one of the forums where they take up ripped and compressed DVD's.

As for dSCaler you can look at http://www.dscaler.org, but that's not really necessary if you use one of Andy's latest filters since he has put the dScaler sharpen plug-in in those builds. If I remember correctly from the last time I saw anything DiVX you really should use any sharpness filter you can with it, they're very blurry.

bugfixed SSE2 version up report bugs/problems per private message pls

can some1 with a pentium M send me info if this version runs without problems on those cpu's too?

Yes, it does. I'm in bed now, but I have my laptop on my...lap, so I tested it real quick (a Lifebook with a Pentium M 1600). No problems with WinDVD 5. But Pirates of the Carribean looks better on my Barco than on this LCD postage stamp...

Worked great on my P4-Mobile 2.2 @ 2.94! Just watched monster with the fiance, disturbing story especially considering how close to the area I was during that time. CPU utilization was at 60-65% versus what was previously 80-85% so there is deffinately an improvement in performance.

Today I spent several hours testing VMR9, VMR7 and Overlay with various Resize settings including no resize.
Doing these tests reminded me of just how subtle the improvements due to resizing actually are.
Many people will be hard pressed to see any improvement at all, so for those of you with slower PC’s, don’t stress to much about resizing unless you want to use its sharpen function as well.

To make comparisons fair, I used a resize setting that would work smoothly with the most demanding output, VMR9 and Denoise3d. This wound up to be 1440x1728 or 2x Horizontal and 3x Vertical PAL DVD.

I believe Denoise3d to be the most useful function of FFDShow and now that it is easier to run, I recommend everyone use it.
I would prefer to give up resize in order to run Denoise3d every time, it greatly enhances my viewing experience and I don’t want to view DVD’s without it.

Now to the VMR-Overlay comparison using my Radeon 9600 Pro.

Without resizing:

VMR7 looks harsh and grainy with noticeable artifacts. CPU usage is low, but I loose the ability to adjust picture settings inside Zoom Player. I can’t think of a good reason to use this one.

VMR9 gives a noticeably soft image with very low grain and excellent color.

Overlay is in between the others for grain and color is good but color balance is off. I find that reds have to much orange.


With resize set to 1440x1728 the differences between renderers is less noticeable.

VMR7 is still harsh and grainy and I just don’t like it.

Overlay and VMR9 are close for grain and noise with VMR9 still noticeably smoother and less edgy but not by a much. The strange thing is that at this high resize setting; VMR9 does not really look any softer then Overlay which is surprising. Color balance is also better with VMR9. Overlay defiantly shows to much orange on my system to be considered accurate. This is quite subtle and is not always noticeable and can probably be compensated for in display calibration.
With the Radeon overlay, I cannot really say that I have a problem with color, but if I had to pick a winner it would be VMR9. It just looks a little more natural.

I played with AA and AF settings in the Direct3d page of the Radeon drivers but I can’t say that I could notice much difference.
There is a note on the page that states that Max resolution is 1280x1024 for AA or AF. Maybe they are disabled at high resolutions.

So in the end, I have been swayed back to using VMR9 over Overlay but I had to give up a little resize to do so.
Without the speed gains from the optimized Denoise3d I think I would still be using Overlay at a higher resize setting.



Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Owen,

I'll turn off resize and try my current sharpening and 3Ddenoise (vs gradual currently using) and see if its better than a small amount of resize and gradual.

Will let you know what i think.

Oh yeah 2400+ that WAS overclocked, but it looks like giving me BSOD Graphics card infinite loop errors, pitty gave me near 15% cpu to play with in FFDShow :(

When I check FFDShow as the secondary post processor in FWMM it messes AR completely, forcing a 4:3 AR, compressing image horizontally even in fullscreen sources, this effect is even worse with anamorphic sources. This happens independent of options checked in FFDShow. Unchecking it fixes the problem.
Using the FWMM filters in ZP this doesn't happen, AR is correct.
¿Any ideas anyone?

Andy,

Since your changes are not going back into the SourceForge tree, how are you accommodating requests for a full source tree of 'your' version? Do you have a CVS compatible site?

Thanks,

Originally posted by MarkStega
Andy,

Since your changes are not going back into the SourceForge tree, how are you accommodating requests for a full source tree of 'your' version? Do you have a CVS compatible site?

Thanks,

I will try to contact milan how to handle this, first i want to do some more work (trying to work on the resizer now)
But might be a good idea to make a new cvs project for the SSE2 stuff.

Andy,

Thanks for the quick reply. Once you resolve how you two are going to update source, can you please post the information here?

No rush, I am really just trying to read the code to understand how FFDshow fits into the video processing schema from DVD to video card. (I'm really trying to understand why TheaterTek has trouble with the resized output from FFDshow post-processing).

Thanks,

Owen, are you applying Denoise before or after resize?

Need some small help, what is the normal color format of a dvd yv12, yuv2 and to what color format u dvd videophile convert?

Im asking since i need to know wich scaling routines i rewrite first. I try to optimize them complete new based on the c code, since if i use the slow c code + lanczos i dont have those horizontal lineing artefacts wich means the bug is only in the mmx version's.

Andy,
when i play a dvd ffdshow info reports yuy2 input color format.

But if i output yuy2 or rgb from ffdsow can't get a proper dve chroma-luminance pattern .
With yv12 it is ok.
I think vpopovic or Li On can be more specific.

As Lanczos resize if i resize using avisynth's lanczos4resize the orizontal line artifacts are gone.

Hope this helps.

Owen,

Thanks for the input. I knew you would come back to VMR one day. In my Nvidia beta system (Quadro with 61.12 drivers, Forceware Multimedia 3.0 Beta, Direct X 9.0c beta) VMR7 and VMR9 produce almost the same image quality (grain, color balance, detail, smoothness), except that VMR9 uses more CPU resources. Too bad your drivers won't let you use AA and AF. Although subtle, I bet you would like the improvement. I am not sure if using FireGL drivers would unlock higher AA and AF capabilities. You might want to check that one.

As far as "orange" effect, that is why I switched to Nvidia. Even with VMR9, reds push towards orange. This also affects entire color balance. That was true using both DVI-D and RGB. I was not able to calibrate RGB gains and offsets on my display to compensate. Somebody suggested that it is "phosphorus level" issue. I do not know all the theory behind phosporus levels (so no guarantees my comments make sense), but I know that with DVI (or at least DVI-D) and Photoshop gamma panal you can actualy adjust phosphorus attributes of your video card to match your display (within the limits of your hardware/driver). Probably another way to do it is to use Powerstrip and choose "match EDID properties of display". However, this option will change gamma curve and tempreture of white as well, while Photoshop will let you adjust all atributes separately. I tried playing with all these settings, but at the end of the day default settings on my Quadro produce a better result (i.e. unmatched gamma and phosphorus level, 6500K white on both display and video card).

Originally posted by AndyIEG
Need some small help, what is the normal color format of a dvd yv12, yuv2 and to what color format u dvd videophile convert?

Im asking since i need to know wich scaling routines i rewrite first. I try to optimize them complete new based on the c code, since if i use the slow c code + lanczos i dont have those horizontal lineing artefacts wich means the bug is only in the mmx version's.

Andy,

To the best of my knowledge software decoders will output in YUY2 format to FFDShow. In default settings FFDShow will convert YUY2 to YV12 before processing the image. YV12 is a faster colorspace as chroma is downsampled (YUY2 has 4:2:2 while YV12 is 4:2:0). Although this causes another colorspace conversion, it is a faster way and it produces more accurate chroma upsampling results as chroma is then upsampled in hardware.

Forceware Multimedia 3.0 Beta (which will become very popular due to its native FFDShow support with the folks that don't want to mess with ZP) can also decode in UYVY format (similar 4:2:2 colorspace as YUY2 with bits organized a bit differently) with Nvidia Post Processing option checked. However, in this case, Nvidia will output YV12 to FFDShow. I heard that new TT will also output YV12 to FFDShow.

So, in short, you should make it work for both YUY2 and YV12. As far as scaling algorithms, I believe most people use Lanczos and bicubic. Spline theoreticaly produces best results of all "ringing" algorithms, so it would be nice to have that one fixed. When you are messing with resize, I second to Owen's suggestion to expand resize beyond current 2000x2000 limit.

Thanks again for the incredible work.

Not sure if this has been discussed, but resize to 1440x960 and lancos causes me a lot of stutters, UNLESS I use the denoise filter prior to the resize. If I do resize only, I can't handle it. Anyone know why that is? The image looks slightly better doing the denoise after the resize, but not substantially, so I'm keeping it before. I almost have enough horsepower to even do a denoise3d HQ before a 1440x960 resize, but not quite, it spikes up to 100% occasionally and I get a bit of stutter, so I still have it on gradual denoise right now. It just doesn't make a lot of sense. Whats magical about doing a denoise prior to resize that makes the whole process so much more efficient?

Denoise works on a frame size of 720x480 pixels (raw NTSC video from the DVD) when it is before the resize & you are asking it to work on frames of 1440 x 960 pixels (4 times the data) if you place it after the resize.

You could experiment with smaller resize values and subsequent denoise to see if you prefer the output (and, of course, if you have the CPU power to avoid stutters).

I have tried a resize to 1280x720 with the denoise after & haven't really decided if I prefer that or the denoise followed by a 1440x960 resize.

He's saying that without Denoise, he can't resize. Not that he can't denoise after he resizes. Or so I read it ;)

Originally posted by Jeraden
Not sure if this has been discussed, but resize to 1440x960 and lancos causes me a lot of stutters, UNLESS I use the denoise filter prior to the resize. If I do resize only, I can't handle it. Anyone know why that is? The image looks slightly better doing the denoise after the resize, but not substantially, so I'm keeping it before. I almost have enough horsepower to even do a denoise3d HQ before a 1440x960 resize, but not quite, it spikes up to 100% occasionally and I get a bit of stutter, so I still have it on gradual denoise right now. It just doesn't make a lot of sense. Whats magical about doing a denoise prior to resize that makes the whole process so much more efficient?

im not 100% sure but unalligned data access (read/writes) and cache hits/misses could cause this. The libav and mplayer dll compiled with gcc result in quite a few misalligned data accesses wich cause extra cpu cycles. If u use a filter before resize the scale routines might still have some cache lines in L2 cache rather than memory or teh scaler takes the data's out of some better alligned buffer and without a filter before teh scaler use the misalligned main buffer.

Theoreticaly every filter could solve this problem before the resize if the resulted datas are written to a better alligned buffer.

madpoet,

I see what you read & I guess I need Jeraden to restate the question:D

I can't explain why a resize alone would stutter, but a denoise follwoed by the same resize would not, if indeed, that was the question.

Originally posted by madpoet
He's saying that without Denoise, he can't resize. Not that he can't denoise after he resizes. Or so I read it ;)

Yeah, thats what I'm saying.

If you don't like the visual effect of denoise prior to resize, you can turn denoise on, but pick the minimum possible denoise setting so it really doesn't alter the image at all, but the side effect is that resize is more efficient. However, that means you can't use denoise again afterwards since you already used the filter.

I'll play around with it tonight and see if I can find another filter which produces the same phenomenon. Hopefully one of those other filters that nobody ever uses does the same thing. :) That way you can still use denoise afterwards and not have to waste it before.

I've been spending the last few weeks playing with ZP, ffdshow, with Sonic filters and have a few observations to make:

ffdshow with the SSE02 mod runs great (noticeably better than any other version I've tried) with denoise 3d, Resize to 1440x960 and VMR9. If it wasn't for the noticeable horizontal tearing, this would be a great combination. Could this be a 9600 non-pro VMR9 issue?

In searching for a workable configuration, I've also tried the same settings using Overlay. With overlay, zp usually freezes on startup, requiring a Task panel shutdown of zp. Most times it immediately freezes showing the zoomplayer splash screen. Other times it may switch to a black screen. I've turned off VMR in sonic applet, tried several versions of zp (including RC2), and different versions of ffdshow (0520 and SSE02), re-installed zp and ffdshow, all with the same result. I'm using zp install configuration defaults for everything except for the specification of ffdshow, and sonic video/audio as filters. With ffdshow removed from the filter list, everything starts up fine. After trying many times there is the odd occasion where it starts succesfully and runs. Then if you stop it and restart, it would fail again. I believe most successful starts follow a restart of the PC.

When the overlay does work I find the image not as pleasing as VMR9, image observations are consistent with Owen's comments.

Thanks,

Peter
------------
P4 3.0c @ 3.2
Sapphire 9600 256MB
Asus P4C800 E

Originally posted by Jeraden
Not sure if this has been discussed, but resize to 1440x960 and lancos causes me a lot of stutters, UNLESS I use the denoise filter prior to the resize. If I do resize only, I can't handle it. Anyone know why that is? The image looks slightly better doing the denoise after the resize, but not substantially, so I'm keeping it before. I almost have enough horsepower to even do a denoise3d HQ before a 1440x960 resize, but not quite, it spikes up to 100% occasionally and I get a bit of stutter, so I still have it on gradual denoise right now. It just doesn't make a lot of sense. Whats magical about doing a denoise prior to resize that makes the whole process so much more efficient?

Somewhere back in this thread I'm pretty sure this phenomena was discussed. I know I had noted it at one point.

Andy's cache/buffer alignment explanation makes sense.

Interesting... so we would use a useless filter in front, then resize, then Denoise3d. Hrm...

Well with the latest SSE2 version I am now able to use denoise3d HQ before 1440x960 and it is hovering at around 85-90% with peaks just below 100%. Picture improvement is noticeable, thanks again Andy.

Owen thank you for that report. It sure comes in handy. Since I have only an AMD 1000 with Radeon 9200se I might just upgrade to my AMD 1800 laying around and simply use Denoise3D for awhile. Since you say there isn't much noticable differant with or without Resize. Denoise3D being the most visiable differance in image quality. I'm sure Resize would make is a little better but not worth spending $300 on another cpu/mb/memory combo.

Since I only have a Radeon 9200se I was thinking of droping in my Radeon 9600se from my other system. These SE's are very low powered compared to Pro's and XI's versions of Radeons. But I thought only effected gaming.
Basement and a few others mentioned that their may be an image quality issue with the SE's? Could that be true?

I use Theater Tek. Which I assume I use Overlay. Should I look at VMR9 like the rest of you? Does TT allow me too? I don't even know where to get VMR9.

Originally posted by stylinlp
I use Theater Tek. Which I assume I use Overlay. Should I look at VMR9 like the rest of you? Does TT allow me too? I don't even know where to get VMR9.

TT only uses the overlay right now. Not sure if it will allow VMR in the 2.0 version.

Originally posted by Eiffel
On my Celeron overclocked to 2.66GHz, I can run denoise3d followed by resize at 1440x960 on NTSC DVDs using Owen's suggested settings. I'm still a little bit low on power to resize PAL discs to 1440x1152 (resizing to 1024x768 does work well with an average CPU utilization of 80%)... but this may be asking too much

Thanks again

Eiffel

How is that possible with a Celeron? I can't do either 1440x1152 or 1440x960 with my P4 2,4@2,6GhZ(533), 512MB.

Another thing, I can't find where to ativate Denoise3D?!

Why can't I just configure "raw-filter only"(doesn't work), but only the regular config?

Is it worth spening a whole lot of cash getting 2x PAL for 3,0@3,6Ghz ++?

stylinlp,

The 9600 non-pro I'm using is not an SE card either.

The tearing is most evident on fast moving scenes or horizontal pans. On those occasions where I can get the overlay to work the horizontal tearing is just not there. For me, it's a toss-up on whether I prefer VMR9 with tearing or no tearing with Overlay.

I just tried the ffdshow-20040607a_P4_AMD64.exe version of ffdshow, but TT crashed all over the place -- always on exit. I was hoping this would take care of my issue. I guess I will go back to the older version.

1. Denoise3D = 0.5, 0.5, 5, & HQ
2. Resize = 1440x960 with Lanczos@10 and L&C each at 1.5

My machine:
MB: Intel D865PERL w/865PE chipset
CPU: Intel P4 2.8CGHz with HT on
Mem: 512 Meg (Kingston DDRAM 256MX2 DDR400 C3 PC3200 K RTL)
Audio: M-Audio AP2496 with v27 drivers
DVD Player: TheaterTek 1.5.64
Video: ATI Radeon 9600 (non-pro) with CAT 4.5 drivers
OS: WinXP Pro SP1 with all latest patches

godavego,

You are using the wrong version. That version did cause crashes on exit from TT. Andy found some bugs and fixed them. The new version:

ffdshow-20040609b SSE2.exe

seems to be bug free (at least with one day of testing on my system).

Try that one.

BTW. I have nearly the identical setup as you do, with nearly the same filter settings. I did an experiment yesterday with hyperthreading. The playback was slightly more efficient with hyperthreading turned OFF. I tested this because it was suggested earlier in this thread. The difference is minor, but measurable.

Hope this helps.

Phil

godavego & phil,

Since you both have almost identical setups to mine ( I have a Radeon 9600 Pro by Sapphire and a 3.0 Ghz P4 with HT off) using the recent FFDshow with resizing and HQ denoise:

Have you found that a single adjustment in TT aspect ratio for each of letterbox, 16:9 & 4:3 mastered DVD's is sufficient? (So far, I have)

Have either of you a totally coherent explanation as to why one needs to make AR adjustments in TT after using FFDshow?

The best explanation given to me (so far) is that TT actually processes the video stream AFTER FFDshow when deciding how to scale each DVD type to the current output resolution and that it does not perform that calculation correctly for other than the raw DVD's 720 x 480 NTSC video stream (or 720 x 525 PAL).

I.E., processing is 1) TT reads MPEG data & creates 720x480 video stream, 2) FFDshow post processes video stream (and in our cases creates a 1440 x960 video stream), 3) TT processes that stream and tries to scale the stream appropriately based upon input AR (4:3, 16:9, letterbox) and output AR (typically either 4:3 or 16:9).

Right after I posted, I found the new version. I'm trying it now with relatively good results. Even video-based content works without dropped frames. Not bad so far.

I'm still struggling with aspect ratio correction though. It's a pain to keep reseting things to make the aspect ratio correct. I may end up having to drop ffdshow because right now it doesn't pass the WAF. For that, I need to be able to pop a DVD in and go. No tweaking allowed.

What can I do to stop ffdshow from messing with the aspect ratios?

There is the option under editting Aspect Ratio that says lock geometery I believe. I unchecked that and it stretched the the image to a 16x9 frame, this however produced the correct geometery, verified with overscan/geometery tests on AVIA, DVE and HTT. Make sure to do that for 16x9 or 4x3 and not the this dvd option.

how much improvement (CPU usage) are you guys seeing with the SSE2 optimized code? I'd love to switch but I am a bit skiddish since it work decent right now with 20040329 on a P4 2.4Ghz (not overclocked) resizing to 1776x1000 with c and l sharp at 1.5 or so and some gradual denoise...no spikes. Should I just hold onto a good thing or try the new stuff?

godavego,

I believe (but I haven't had enough DVD's passed through the system to verify) that you only have to make 3 aspect ratio settings once in order to get proper AR. Unfortunately, I stumbled on this right before a business trip & won't be able to confirm until this weekend.

1) For each DVD type (4:3, 16:9, letterbox), it appears that TT sees the flag (if you bring up the AR editor in TT, the appropriate AR is in blue.

2) You have to pick the selected AR (it defaults to current disk)

3) Uncheck "keep ar" (by memory, might be "lock ar")

4) Adjust for proper AR

5) Save

So far, having three settings seems to be working consistently. I do not understand what TT is doing that this is wrong & I don't understand if I am adding yet another scaling step from TT. But, the output still looks better than without FFDshow.

This weekend, I'll pull out my DVE to make certain that I am truely getting a measurably correct AR.

Mark,

With previous versions of ffdshow, I have found that a single master setting for each AR works well. I have encountered problems with very few DVDs, and in those cases I suspect the DVD is flagged incorrectly. I haven't had a chance to extensively test yesterday's version, so I can't really comment, but so far so good. As to why TT behaves the way it does, I don't know.

godavego,

Maybe you are doing this already, but if not, it may be the source of your AR problems. In TT, when you activate the AR editor, you must select the specific AR and then make your edits. If you just make your edits with the top selection highlighted (I forget what it says), you are only editing the AR for that specific DVD.

So far I am not experiencing any AR problems following the initial setup.

Hope this helps,
Phil

Phil,

In my short series of tests, I agree, it appears that disks correctly encoded only need the one setting (of course, they were 'wrong' also under plain old TT (or a hardware DVD player for that matter)). Using the AR editor to fix 'current DVD solves that problem.

Dan,

I haven't objectively measured, but there were too many stutters at 1440 x 960 before this version and I only resized to 1280 x 720, now they are insignificant at 1440 (and the PQ appears better to my eye).

Originally posted by debennett2
how much improvement (CPU usage) are you guys seeing with the SSE2 optimized code? I'd love to switch but I am a bit skiddish since it work decent right now with 20040329 on a P4 2.4Ghz (not overclocked) resizing to 1776x1000 with c and l sharp at 1.5 or so and some gradual denoise...no spikes. Should I just hold onto a good thing or try the new stuff?

wait a week im working on the resizer atm, if this is done the new version should free 10-30% cpu usage depends on filter combo. There may be some little bugs to smash before all works like usual. The actual versions are all kinda test version but so far no new bugs are reportet for the latest version.

Do any of these latest versions after the 5/25 build have any benefit for non-SSE2 processors?

Originally posted by Jeraden
Do any of these latest versions after the 5/25 build have any benefit for non-SSE2 processors?

Yeah, the non-sse2 build is optimized for mmx2. Makes a decent difference on my AthlonXP system. (10-15% cpu on HQ Denoise3d)

godavego, doesn't TT have a DVD setup memory like Zoom has? I put in every new DVD once, select the correct size and placement (I move the image to the top of the screen since I sit on a plattform), the correct soundtrack (english on most, norwegian on children's DVDs) and subtitles (norwegian). Finally i start playing and hit a button on my remote, which makes Zoom create an auto load bookmark with all the correct info, and it starts at the beginning of the movie, not with any commercials or long winded menues. So when the wife or kids wants to watch a movie they just pop it in, and everything is correct. :cool:

Andy, is it possible to make the OSD less hungry? I scared myself to death this morning, I suddenly had wild stutters and a maxed out CPU even without denoise3D! Turned out it was because I had increased the size of the font in the OSD, and it created a huge load on my CPU!

ANyway, 10-30 % less CPU use on resize sounds fantastic! Can it mean that triple resolution is within reach? I've got my 3.0 up to 3.9 now, and I believe it's stable on that, so the horsepower should be there (I got it up to 4.0 just for the bragging rights, but it wasn't Prime stable, so I went down again a bit - which is why the motherboard brand is called Abit - you go "up abit and down abit" ;) ).

Originally posted by madpoet
Owen, are you applying Denoise before or after resize?

Before resize.

I am not yet convinced that running Denoise3d after resize offers any advantage and at the time of my last testing, 12 hours ago, it was still way to slow.

This could be related to the problem of slow resize performance without a noise filter being used before resize. I can confirm that I have noted that problem with the last few builds. I found the older Athos builds were ok in that regard, although others report having this problem for quit some time.

When I can get it running smoothly with generous resize I will re evaluate.


For those of you still using Gradual denoise, think again.
Denoise3d used before resize far outperforms Gradual denoise used anywhere.

Recommended settings for Denosie3d BEFORE resize are:

L 0.5, C 1.0, T 5.0 HQ

These settings do NOT remove detail and DO remove a lot of noise without any visible side affects.
It’s boody miraculous. :D

Gradual denoise has nasty motion artifacts at any useful setting even after resize.


Regards,

Owen

I don't want to post in the wrong thread, so could someone PM or email me and direct me to the right place to post problems? This is my first time trying out ffdshow with ZP, and I want to have the best possible picture, of course.

TIA!

Originally posted by stylinlp
Owen thank you for that report. It sure comes in handy. Since I have only an AMD 1000 with Radeon 9200se I might just upgrade to my AMD 1800 laying around and simply use Denoise3D for awhile. Since you say there isn't much noticable differant with or without Resize. Denoise3D being the most visiable differance in image quality. I'm sure Resize would make is a little better but not worth spending $300 on another cpu/mb/memory combo.

Since I only have a Radeon 9200se I was thinking of droping in my Radeon 9600se from my other system. These SE's are very low powered compared to Pro's and XI's versions of Radeons. But I thought only effected gaming.
Basement and a few others mentioned that their may be an image quality issue with the SE's? Could that be true?

I use Theater Tek. Which I assume I use Overlay. Should I look at VMR9 like the rest of you? Does TT allow me too? I don't even know where to get VMR9.


Yes, resizing on its own does not improve quality all that much but resize sharpening is still the best sharpening option IMHO.
So for those of you who need or want sharpening, resizing becomes essential, only to get good quality sharpening.

I don’t believe that the 9600SE cards should produce quality any different to 9600, 9600Pro cards, but VMR9 operation is almost certain to be problematic.
Because the 9600SE is an El Cheapo card, the manufacturer my have cut corners with the card design to cut costs. That would affect quality.

Regards,

Owen

Originally posted by Mastiff
Andy, is it possible to make the OSD less hungry? I scared myself to death this morning, I suddenly had wild stutters and a maxed out CPU even without denoise3D! Turned out it was because I had increased the size of the font in the OSD, and it created a huge load on my CPU!

ANyway, 10-30 % less CPU use on resize sounds fantastic! Can it mean that triple resolution is within reach? I've got my 3.0 up to 3.9 now, and I believe it's stable on that, so the horsepower should be there (I got it up to 4.0 just for the bragging rights, but it wasn't Prime stable, so I went down again a bit - which is why the motherboard brand is called Abit - you go "up abit and down abit" ;) ).

Come on Mastiff, just disable the OSD and be happy. :D

Personally, I have no interest in CPU load as long as playback is smooth.
I don’t need a CPU meter to tell me that the system is overloaded.

You should already be able to get 3x DVD resize.
I can easily run Denoise3d followed by Lanczos resize 1920x1728 with Overlay.
That’s the highest clean scaling resize under the 2000x2000 limit.
For VMR9 I am limited to 1440x1728.

Is your system running 1:1 or 5:4 memory timing?
I am using 1:1 on a 1Gig FSB with memory at 250Mhz 2.5,4,4,7.


Regards,

Owen

Thank you for the feed back Owen.

I spent 5 hours today changing out my AMD 1000 w/MB for my AMD 1800 w/MB and re-installed WinXP Pro.
Got the newest versions of TT, ffdshow and Catalyst. Windows updates.

Im happy. I get no skips at all and the image quality with just your Denoise3D settings look great! Now Im looking at Unity, %50 or %100 gain on my Extron unit and wondering where I like it at. :)
15' BNC cable from Tim.

Need to decide between 1280x720 and the 1440x960 I have it at now. Working on that next. THanks once again

Mastiff,

If using VMR9 and Zoom just right click the screen (during playback), click on Filter Properties and then right click on the VMR9 filter and you will get very nice FPS and jitter values. The FPS resets with any pause. It has no CPU load that I can see and seems very accurate. The jitter value correlates well with my sense of visual jitter and creeps up from 2 to over 30 as CPU load increases.

By the way... congrats on your new system!

Charlie

Owen, it's running 1:1. I believe I'm able to get it stable on 260 FSB, but I need a bit more Vcore. I'm going to try that today.

Charlie, thanks! I'm very pleased with it so far. Let's see what happens when I hook it up to the Barco! I'm using overlay since it's a Radeon card, but I will test VMR9 as well, and I'll check out that property page. :cool:

I have two issues with my HTPC that I think are somehow related. Could someone in the know give me their take on this situation?

HTPC is a 2.8Ghz P4 Intel 865PERL, 1Gb RAM, XP Pro, HT enabled, overlay mode, NVidia FX5700U video, latest ZP Pro, WinDVD video decoder. I run only Levels (first) followed by resize to 1440x960, lanczos or bicubic (usually lanczos). With this everything is fine. But it seems that if I add any additional filters (pre or post resize) I get odd stutters and jerky pans. denoise3d ahead of resize, or dscaler sharpening before or after resize all produce the same results.

The second issue is when switching from Overlay to VMR9 windowed. With ffdshow set only to resize (same settings as above) and my Levels usage disabled, the scene movement looks like it's crawling from top to bottom during screen updates, and video tends to lag audio.

Both of these issues sound like not enough processor power, though TM shows floating around 60%, and ffdshow OSD shows it bouncing between 50 and 90 or so. reclock is used, and anydvd is active. ffdshows tried (no difference) were the latest one 20040520 and the one right before it (20040415 I think). It sure seems like 2.8Ghz should be plenty for this, so maybe there's something else at play?

Any advice?

--Bill

I'd say that resize does matter a great deal, at least on 1280x720 DLP. With progressive resolution you get to see everything loud and clear. If I had to choose one thing in FFDShow, it would be resize. In fact that is the only thing that I am using. I don't care to much about the noise as in the theater I like to sit at distance where noise is visible. With perfect resize and AA and AF there is very little noise to start with, although I acknowledge that very few setups can do both at such high resolutions, so for the most part denoise is probably necessary.

Perfect scalling for 1280x720 output resolution IMO is Avisynth Lancsoz or Lanczos4 960x540 followed by FFDShow bicubic or Lanczos at 1920x1080 (exactly 2x Avisynth resize, so no bugs there). Then card scales it down to output resolution of 1280x720 (divides by 1.5, also nice number for scaling). As perfect as it gets, short of being able to do 1920x1080 (or even better 2560x1440, 2xoutput) all the way in Avisynth. I added some more cooling to my AMD FX-51 (clocked @233 Mhz 2:1 RAM @ 466 Mhz, almost 2.6 Ghz on CPU, 930 mhz on HT, running 27-29 C under full load) and Nvidia Quadro FX 1100 and now I can finaly do all this with Forceware Multimedia 3.0 Beta, VMR7, followed by 8xAA and 8xAF processing.

With perfect scaling and AA+AF good superbit NTSC 740x480 transfers are darn close to broadcast HD, at least in my setup. I would love to post some recent screenshots from Spiderman Superbit in 1280x720, so if anybody has any ideas where could I post such a large image (so at least owners of 720p displays can see it native without scaling), please let me know. Obviously I would post plain 1280x720 image from software decoder for comparison. At recent NY Home Theater Expo I saw a lot of nice scalers, and I can say this image easy compares with 10K Faroudja DVD/Scaler combo, except in smoothness (3:2 pulldown) and deinterlacing (darn combing happens every now and then).

Andy,

Thanks for your great effort! The latest makes it possible for my system (P4 3.0e at 3.6Ghz, Zoom Player, WinDVD decoders) to do 1440x720 with denoise3d, Picture properties offset, Levels rescaling and gamma with YUY2 out. FPS drops to 23.87 and frame jitter is about 13ms which is quite good especially since my O/S has a ton of stuff running! I get even better results on a newly installed O/S but the AbstractDMO filter is not set up properly so the video is amusing to say the least.

IV12 is the speed champ with CPU utilization from 85% to 95% and looks very good.

Disappointingly, RGB32 output is slow and RGB24 had some kind of problem.

Another interesting thing is that video will run for short periods without the AbstractDMO filter. It always used to freeze immediately. I have changed more than ffdshow so who knows what is helping but it seems likely that the Zoom/ffdshow interface issue is nearly fixed.

It's great to be able to use YUY2 for experimentation and comparison with YV12 but that will have to wait until tomorrow.

Charlie

Owen

I mentioned earlier i'd give denoise3D at try at the expense of resize.

Thus far im pleased. I was only resizing to 1024 x 768, followed by some "asharp" and then "gradual denoise". I had some CPU to play with but need such as i watch DTV captures with FFDShow while also capture 2 channels at once (Dvico and Twinhan cards) so maxing the CPU is bad mmmkay :)

Anyway i cancelled resize (note desktop is 1024 x 576) replaced gradual with 3D and the improvement is much better. There are far less "ringing" artifacts. I feel i can extra just as much detail, but "distance" shots now look somewhat smoother, not to mention that the denoise function, while some times "suttle" does a good of background noise without stripping the detail of things in focus etc.

Only think i don't understand is WHY i was getting more artifacts (ringing etc) with the resize than without? Even if i went i higher resize? I was also resizing and sharpening (with full settings) hmmm perhaps this is it?

One last thing, with the IV12 and YUY2 outputs, which is better (yet to to much test). YUY2 does use noticably more CPU.

Ahh display is infocus X1 (800 x 450 on 16:9 screen) feed 1024 x 576.
Also testing settings on CRT 17inch flat at 1280 x 1024

Holy cr...! I really am Superman! Right now I'm running a totally stable (not Prime95 stable, but totally stable in DVD playback) 3.9 gHz on my 3.0, CPU temp only 49 C (probably around 43-44 because of Max3's error in reporting. With this I have first dScaler sharpness, then Denoise 3D and finally a resize to 1920x1728 Lanczos! CPU use is between 80 and 86 %, and pans are rock solid. I tried running this with VMR9, and it didn't report any errors. But the image looked like it was painted by a modern artist! How come? It looked like a small portion of the picture was stretched to cover the whol width, and then the paint was slightly stirred on it. I can do a screenshot, but I'm sure somebody here can tell me what this is. Anyway, on Radeon overlay this is very promising. I hope I'll have time to connect it to the Barco this weekend! I must find the time! :D

Hey Owen, I forget... are you using Avisynth Lanczos or just the one in FFDShow? Also, which codec are you on these days? Still Elecard?

Mastiff, regarding setting up the AR for DVDs before my wife wants to watch, this may not work too well for me. While I do have a modest DVD collection (~100), the majority of the DVDs that we watch are from Netflix. With 13 month old twins, our movie viewing time is limited, so lately we've been sticking to movies we haven't seen yet. I'm pretty particular about which movies I buy vs. rent.

I did figure out the "lock AR" setting for most of the DVDs. What I need to do now is load up my DVE and make sure I have the popular ARs properly setup. I didn't run into this much until I started playing with ffdshow. With TT only, I had things pretty close to an STB. Ffdshow set me back a bit, but the pq is pretty sweet.

Bill,

You might try turning off hyper-threading in the bios.

Charlie

Godavego,

I ran into the same problem and used the "lock ar" to fix it, however any dvd you have previously played will still have the ar memory and you will have to update that particular dvd. I then put in a new dvd that has not been played before and it automatically corrected the AR.

Alrighty, well I guess this is the thread to post my issues:

I tried to do the basic setup, like the FAQ on HTPCnews suggests. I got the Elecard player/codecs in addition to WinDVD 6. It worked for a little bit using both WinDVD video/audio, but now with FFDshow enabled, ZP tries to access the DVD menu, but then kicks back to the ZP splash screen within 2 seconds.

For file playback, like an xvid clip, the audio lags behind the video horribly, but the video playback isn't stuttering or anything.

The system is as follows:

Antec Aria case
nForce 2 mATX board with voltage adjustment
AMD XP 2600+ mobile Barton, stock
Vantec AeroFlow
Hercules Game Theater XP
2x 256 PMI 2-3-2-6 DDR400
Western Digital 100 and 160 gig HDs
NEC DVD burner
ATi Radeon AIW 8500DV

With this knowledge, is there a set of FFDshow settings I should use? I have a very good background with computers, but this software is completely new to me...a little boost at the beginning is much appreciated. :D

With Andy's new FFDShow 20040610b (or something like that) on my P4 2.8C, 800FSB, 512MB 400MHz DDR, I can run the following, and it looks spectacular:

VMR9 with Elecard 2510 on ZP
1. Denoise3D at 0.5, 1.0, 5, HQ
2. Built-in Dscaler sharpen @ 64
3. Lanczos resize to 1440x960 with Luma@1.3 & Chroma@0.5

Perfect picture, smooth video. WinDVD5 with VMR9 is very good with this same setup also.

I really think a big difference in what you can do in FFDShow has to do with the FSB you are running. I see lots of P4 3.0GHz's or higher with 566, or lower FSBs that can't do what I can do with FFDShow.

Originally posted by EvilYoda

It worked for a little bit using both WinDVD video/audio, but now with FFDshow enabled, ZP tries to access the DVD menu, but then kicks back to the ZP splash screen within 2 seconds.



I had the same problem with 1 specific dvd. I changed the DVD i was using and it instantly worked. It was LOTR:TT Extended. All other dvd's tested, thus far, have been fine -- maybe give that a try?

It's not a fix, however, for that specific DVD, so if anyone else has any other ideas to what could be causing this, it'd be great to know.

Greetings all!

I am having trouble using the VMR9 overlay. When I use it, I no longer have any picture controls. The sliders are there, but have no effect. Is there any way around this, or do I have to adjust the contrast, brightness, etc. in another way?

Thanks in advance, and thanks to all of you guys for this great thread!

Jerry

Originally posted by KingKong954
I had the same problem with 1 specific dvd. I changed the DVD i was using and it instantly worked. It was LOTR:TT Extended. All other dvd's tested, thus far, have been fine -- maybe give that a try?

It's not a fix, however, for that specific DVD, so if anyone else has any other ideas to what could be causing this, it'd be great to know.

Hmmm, I'll definetely try another DVD. I had Lost In Translation laying around, so let's hope it was the DVD. :(

Anybody else with a similar system to mine? Could you PM me with which settings you're running?

I know Owen and maybe some others suggested not using Chroma sharpening at all. I can't tell a difference between Chroma at 0 and Chroma at 2. Luma seems to be the only thing that does anything. For people using Chroma, is it because you actually see a difference or just because its the consensus that you should use it? What is it supposed to do anyways? I can't see it doing anything.

Originally posted by Jeraden
I know Owen and maybe some others suggested not using Chroma sharpening at all. I can't tell a difference between Chroma at 0 and Chroma at 2. Luma seems to be the only thing that does anything. For people using Chroma, is it because you actually see a difference or just because its the consensus that you should use it? What is it supposed to do anyways? I can't see it doing anything.

I did some testing last night regarding this. When using only Chroma sharpen set to 1.3 I hardly saw any difference from the original picture.
When increasing it to 1.66 I saw a little more detail. However, using only Luma at 1.3 did a much better job. So yes, there seems to be no point using Chroma when resize-sharpening.

Originally posted by Jeraden
For people using Chroma, is it because you actually see a difference or just because its the consensus that you should use it? What is it supposed to do anyways? I can't see it doing anything.
When I tried it months ago, all chroma sharpening did was highlight edge enhancement without bringing out detail, whereas luma did both. I've had it set at zero ever since, and luma at 1.2-1.4. You can try it for yourself, just watch the image as you move the slider -- the sharpening effect isn't subtle.

JoeFigueiredo,

Which graphic card are you using? The only problem I'm having with zp, sonci, ffdshow, and VMR9 is the tearing. If I can get rid of that I'll be happy. My hardware set up is very similar to yours. I tried elecard 2510 once but didn't spend too much time with it. I didn't look for tearing artifacts but there were a few issues. I had to do an aspect ratio configuration change. Fast forward/rew in zp doesn't appear to refresh the display properly to let you see where it is and it wouldn't allow playback of DVD as-is without something to do the decryption. I imagine all of these are correctable. The display quality was great. I believe that overlay worked consistently as well in this configuration unlike whatever is ailing my zp, ffdshow, sonic, overlay configuration.

Thanks

ProjectorGuy1,

I think I noticed the same thing on older system I have. It's a P3, running a really old version of zp, and the Cyberlink filter on an Radeon 9000. I wonder if you might have to upgrade to later versions of software.

Hey Guys,
Does ffdshow work with Sonic Cineplayer? I've tried just using those two programs and it's not doing anything the to playback. Like the FAQ described. I went in my registry under the sonic section and created a new Dword entry. Then I set ffdshow to the settings the FAQ described and nothing is happening. What am I missing? Do I need specific versions of the programs? I don't see anything in the config menu's about using ffdshow. Thanx for the help.

Originally posted by basement
JoeFigueiredo,

Which graphic card are you using? The only problem I'm having with zp, sonci, ffdshow, and VMR9 is the tearing. If I can get rid of that I'll be happy. My hardware set up is very similar to yours. I tried elecard 2510 once but didn't spend too much time with it. I didn't look for tearing artifacts but there were a few issues. I had to do an aspect ratio configuration change. Fast forward/rew in zp doesn't appear to refresh the display properly to let you see where it is and it wouldn't allow playback of DVD as-is without something to do the decryption. I imagine all of these are correctable. The display quality was great. I believe that overlay worked consistently as well in this configuration unlike whatever is ailing my zp, ffdshow, sonic, overlay configuration.

Thanks

I'm using an AOpen (Nvidia) fx5900xt. Apprently with this card you need to go with VMR9 rather, than Overlay, in order to utilize the card's 128-bit floating point whatever.

I used Sonics as well in the same configuration, but found them too grainy in comparison to WinDVD and/or Elecard. Never had any tearing with VMR9.

Is tearing what I described on the previous page, that the image looks like it's been painted by a modern artist, with parts of it swirled around and other parts stretched?

Originally posted by Projectorguy1
Greetings all!

I am having trouble using the VMR9 overlay. When I use it, I no longer have any picture controls. The sliders are there, but have no effect. Is there any way around this, or do I have to adjust the contrast, brightness, etc. in another way?

Thanks in advance, and thanks to all of you guys for this great thread!

Jerry

Which player are you using?

VMR9 and overlay don't happen together. It's one or the other. Assuming you mean VMR9 and not overlay, your player is probably communicating with the overlay controls.

VMR9 is subject to the desktop (i.e. normal Windows) controls. Check the properties for your display adapter (but not the ones that are labelled 'overlay' -- they're probably labelled 'desktop').

Sorry I didn't state that. I am using Zoomplayer Pro. When I am using VMR9, I have to unselect Overlay mixer of course. Does Zoomplayer have VMR9 overlay controls, or do I have to use the video card controls?

Alos, Zoomplayer Pro it seems cannot open a ripped file like the older versions. When I point to a folder with the video_TS and audio_TS folders in it, I get a "No DVD found" message. I can still open the file using the "open" command, but I liked the function under "Override DVD Drive/Path" that worked for me before.


Thanks again!

Jerry

Charles Black,

WinDVD filters will work with Zoom player and FFDShow without the Abstract DMO filter if CPU load is low enough and no frames are ever dropped. If a frame is dropped for any reason, video will freeze if the Abstract filter is not used.
Since the new builds of FFDShow use less CPU power, fewer frames are dropped so freezing will be lees common without the Absract filter.



Spoonfed,

My suggestion to use Denoise3d was not just directed at you but to others reading this thread. :D

Quote

“Only think i don't understand is WHY i was getting more artifacts (ringing etc) with the resize than without? Even if i went i higher resize? I was also resizing and sharpening (with full settings) hmmm perhaps this is it?”

Using resize sharpen at 2.0 would explain that problem.
Also note that there is no reason to use Chroma sharpening. It will only increase noise, not detail.
Remember that Chroma in DVD’s is only half the resolution of Lumma.
By all means use the Lumma sharpen, but go easy. Less is more. :D

Quote

“One last thing, with the IV12 and YUY2 outputs, which is better (yet to to much test). YUY2 does use noticably more CPU.”

I recommend YV12 for FFDShow output colour space.



Mastiff,
I have now idea what is happening with that “painted by a modern artist” look.
It is definitely not tearing. You will know tearing when you see it, as it looks like a horizontal tear through the picture with one half of the picture displaced horizontally on fast panning shots.


Madpoet,

I am using FFDShows Lanczos resize at the moment. It allows me to use 1440x1728 with Denoise3d and VMR9.
I could not get near that resize using Avisyth resize.
I will give vpopovic’s strange resize combination a go today and report back.
Elecard 2510 decoder is still my choice. When Forceware 3 is released I will give it a go, but I can’t be bothered with the beta.
I don’t expect FW3 to be any improvement in quality over Elecard based on the posts in the thread I started, but many will find FW3 a better option due to Sub support and better AR for NTSC users. Both issues do not affect me.


Projectorguy1

Quote

“I am having trouble using the VMR9 overlay. When I use it, I no longer have any picture controls. The sliders are there, but have no effect. Is there any way around this, or do I have to adjust the contrast, brightness, etc. in another way?”

Picture controls in Zoom Player with VMR9 output work just fine here on both Radeon and Geforce cards.
I just right click in Zoom Player and select Color controls from the menu. It works perfectly so I don’t understand what is happening with your system.


Regards to all,

Owen

_____________________________
The FFDShow resize – sharpen dude

Owen,

The issue of color space has been a recent thing. Never really saw it mentioned except for the last week or so.

I checked my ffdshow setup and nearly all of the methods are checkmarked. Are you saying none but YV12 should be checked? Is that recommendation for minimizing CPU load or for PQ, or both?

Also, for the resize sharpen settings (Lanczos), I rarely see any mention of the "Parameter" setting. What does it do? What is the recommended setting? I've seen people say theirs is set to 10, but when I set mine to 10, it looked like PQ was degraded.

Thanks,
Phil

Im using the june 3 2002 version, do I need the graphs to use it with zoom player?

Owen,

Unless you finaly decided to upgrade your tubes with some fine pure digital gear, I am not sure if you will find my settings to be better that your current settings. They are optimal for 720p display.

Based on the tweaking experience in my setup, I like to break 4:3 to 16:9 barier in software, preferably Avisynth. Slightly better scaling. Then I really don't like FFDShow horizontal scaling bug. While not visible at its worst all the time, it does create a bit of haze over the image all the time. So that is why first Avisynth for a little resize (960x540), and then 2x resize (2x vertical and horizontal, no bugs) to 1920x1080 with Lanczos. Best scalling, optimal CPU usage. That is a great way up. But the question is now how to go down from here.

The benefit of 1080p software scaling for 720p output is that my card downscales it really well. I believe it likes 1.5 divider for horizontal and vertical resolution. 1080p gives me much better result than 1008p, when you consider total number of pixels both have. For example benefit of using 1008p as opposed to 864p is not as much as 1080p over 1008p, although pixel count difference between 1008p and 864p is double than difference between 1008 and 1080p. Again, it is downsizing divider that appears to have even more impact than resize pixel count.

I think you need to work backwards from your output resolution and use Avisynth to do your first (smallest and trickiest) resize and break aspect ratio barrier. This is based on the theory that you will have to do odd scaling at least once, so you need to throw your best scalling gun at it (IMO Avisynth does it the best). Video card really has the worst scalling and therefore it needs a break with an easy task it can handle well (downsizing by 1.25, 1.5 or 2, from worst to best, I'd avoid odd dividers). Then in between you need clean 2x (for vertical and horizontal) multiplier for FFDShow.

I must admit that for most "normal" people this will not be a big issue, but since this is FFDShow thread, it is probably worth mentioning. And obviously it is not worth hardware upgrade unless it is something you wanted/needed to do anyway.

Owen,

Thanks for the DMO_V Absract details. What you say explains my situation exactly. Do you know what DMO_V Abstract does? I just got mine running again so the question is more for general knowledge than for getting it to work.

Thanks again, Charlie

Charlie

Vpopovic,

Where can we download the Avisynth filter that you use? I remember that Avisynth filter's latest version had compatibility problems when selected within ffdshow.

Thanks for your guidance.
Taci

Owen, are we gunna let that salvo at the glorious Cathode Ray Tube slip by?

That's like waving a red flag at a bull..."fine pure digital"...yeah, righto whatever you reckon Vlad ;-)

@vpopovic "Then I really don't like FFDShow horizontal scaling bug."

I think if the speed of my implementation is good this bug will be gone. Its in the yuv scaler routines wich work with odd rounding offsets to save cpu time in the mmx2 version, but this way they arnt 100% correct thats why the bug occurs.

Im using a total diff. algo to work on the yuv scaling values and it seems that my implementation is more correct and result in the same values than the org. c routine.

To bad i cant yet say if my routine is faster or slower, the org. code is realy tricky and nicly done. I have to finish the final scaling routine. But my SSE2+ final c routine dont have this bug, seems also that all color's are 100% perfect (no green shifting). Im off over the weekend (grilling time) so i will try finish the work next week.

happy weekend all

how do you guys think about the sinc resizer?i think it is the best,the images get
smooth,sharp,and rich color.but it really hungry for cpu usage,my cpu is amd barton 2500(2.2ghz),i can only resize to 960X750 whit blurr&nr on, luma sharpen 0.88,if make resize higher,images will be jumpy.and if turn blurr&nr off,i even can't resize higher than 720X480 with smooth playing because cpu always in 100% loaded.
any comments?

Thanks Andy for your efforts.
Since you're working on the scaler code can you check avisynth 's Lanczos4resize to see if can implement to ffdshow's Lanczos the 8 or better
tap scaling.
Unfortunately i don't know what taps are or why the number of taps affect scaling quality but trying avisynth's LanczosResize which is 6-tap and Lanczos4Resize which is 8-tap i see that there is a "better" result.

Thank you anyway for the great job you've done so far.

btw,where can i down ffdshow that later than 20040520?
thanks

Originally posted by vkon
Thanks Andy for your efforts.
Since you're working on the scaler code can you check avisynth 's Lanczos4resize to see if can implement to ffdshow's Lanczos the 8 or better
tap scaling.
Unfortunately i don't know what taps are or why the number of taps affect scaling quality but trying avisynth's LanczosResize which is 6-tap and Lanczos4Resize which is 8-tap i see that there is a "better" result.

Thank you anyway for the great job you've done so far.

hehe im asking myself the same question's... all i know from the code is that in lanczos mode i have to work on 8 diff. planes per yuv value. The normal bicubic is using 4 planes i think and sometimes the values differ from 2-8, i think this is the tap thingy.

btw there is no diff. using bicubic or spline or lanczos or sinc general code wise. The only thing's wich differs are some coef values and that "tap" thing wich can be from 2-8 it seems. All other is handled the same way, all use the same general resize routine.

I just tested it for the diff. modes

lanczos = 6-8 planes

spline = 10 planes

sinc = 21-23 planes !!!

so we have to work on nearly 3x the values in sinc mode compared to lanczos

woops i just noticed that my routine can max work with 8 planes... damm i have to write a new one for spline and sinc than using a other method...

Right now i'm doing some search in the net and all i understand so far that it's some kind filtering so the upscaled image to keep the original image's sharpness and have as less artifacts as possible.
I found that faroudja's 2300 chip uses 12-tap and terranex scaler 32.

Do you have access to avisynth's lanczosresize and lanczos4resize?

Oops!

I saw some other members posting about spline's and sinc's superiority but since they produced stutter they stopped using them.

I 'll try them tonight to see if they produce better results.

You think you could optimize spline's and sinc's code so we can start using them effectively?

Andy,

Thanks for the great work. Even if your new code does not speed things up, ironing out the resize bug will be a revolution. Don't take my comment as too harsh - I guess it is just a fact.

As far as the colorspace, I thought that FFDShow by default converts YUV2 to YV12 for processing? Isn't that the case? With current code YV12 processing is much faster than YUV2. Are you testing it with YV12 as well? Also, YV12 ultimately results in better chroma as it is correctly upsampled in hardware. Although lot of decoders claim to be chroma bug free, they can't resolve chroma resolution pattern in DVE if YUV2 is passed to video card, while they can in YV12 is passed on.

As far as taps, as I understand it, it is number or surrounding pixels that are taken into account to determine atributes of a resized pixel. The more the better, as resulting image blends better with surrounding pixels. I think Nvidia and ATI use 5 horizontal, 3 vertical tap (8 tap) scaling in their current models.

Teranex with its 32 tap scaling is a monster in a class of its own and 50K is the cheapest Teranex unit, so we can forget about that one. That technology was de-clasified from military technology restricted list couple of years ago, and was developed with even higher demands in mind than videophile use. They wanted evey pixel to be 200% perfect as it (i.e. individual pixel) might represent a nuclear missile on their huge war game screens.

Lancos6 is 12 tap algorithm and at one point it would be interesting to give it a try. There are Lanczos6 algorithms available, but only for photo, not video (to the best of my knowledge). I agree that Spline and synch are promissing, but extremely CPU hungry. The benefit is that with synch I realy don't feel compelled (so far) to scale above the output resolution, but I need to play a bit more with it to evaluate it.

Taci,

You can get 2.5.5 beta (or alpha, I forgot how they call it) version at Avisynth development forum. Do a net search and it will pop up.

Vlad,
When I see a digital RPTV that can outperform a good CRT RPTV I’ll buy it.
So far that has not happened.
To be fair, on this side of the pond we don’t get the range of models available in the US. For example, the Samsung models available here are all LCD based.
Maybe when the second generation 1080p LCOS, DLP etc models become available they will provide me with a worthwhile upgrade path.
Until then, I shall continue to enjoy the complete lack of pixels and deep blacks that only CRT RPTV’s can provide. :D


Regards,

Owen

Owen,

Just kidding. At this point RP CRTs are realy good. They do lack digital DVI punch though and capability to resolve higher progressive resolution. With new Samsung's HD2+ models (as opposed to HD2 that I have) I think most people will agree that DLPs are over CRTs though. Increased contrast ratio and black levels are what makes the "+" difference. With LCDs, I'd agree with you, although 70" GWIII is realy tempting.

Q1 2005 is when 73" 1080p Samsung model is anounced, and I'll make sure to be on the waiting list. I can't wait to put my nose 7 feet from this puppy.

Andy,

Thanks and great work. Now I wish you could get the code for the AC3Filter and work your magic on it. Your (bug and efficiency) polish could make this filter great (and work with my M-Audio 410 card). :)

I'm starting to get a little confused with the colorspace output... have I been missing this setting all along in FFDShow? I don't think I've ever touched anything related to it. Where would I find these settings?

Tested Andy’s 20040610c_SSE2 build today.

This build works very nicely and allows me to run Denoise3d followed by FFDShows Lanczos resize to 1776x1728 with VMR9 which I could not manage before so the speed gains are definitely there.
Great work Andy and keep it up.
I’m sure that everyone here is grateful for you hard work.
If we can get similar speed gains out of changes to the resize code as we did with the changes to Denoise3d we will all be ecstatic. Even people with modest PC’s will be able to get very worthwhile results.
Thanks.


Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Yesterday I noticed a problem on my setup with denoise3d. I'm using TT w/ the latest SSE2 optimized ffdshow.

I have an 852x480 plasma, and on the right side of the screen there is a band of approximately 20 pixels that has a slightly different color. Can't tell if it's different brightness, or what it is, but it's very noticeable in some scenes, and less in others, depending upon the color and brightness of the scene itself.

My default settings are Luma 0.5, Chroma 1.0, Time 5, HQ. I did some brief adjusting this morning and the band disappears if Luma is set to 0, or if Luma is increased to around 1. The band only seems to appear when the Luma setting is between 0.1 and 0.8 (where I want it!).

Has anyone else observed this?

Phil

BTW, I checked all of the other filters that I use (resize/sharpen, levels, picture properties) and the problem is definitely with denoise3d.

Spent more hours today testing every resize combination I could think of.
Nearly drove myself crazy. :D
I cannot come up with a combination of Avisysth and FFdshow resize that works well with my 1776x1000 desktop resolution when scaling up PAL DVD’s.
Nothing I tried was any better then using Avisynth or FFDShow Lanczos resize on there own.

One thing has become clear to me and that is that VMR9 only looks good to me at very high resolutions.
At lower resolutions, VMR9 is way to blurry for my liking.
Without resize, VMR9 looks dreadful even though the color is nice.

With the Andy’s 20040610c_SSE2 build, my system can run Denoise3d followed by Avisynth Lanczos or Lanczos4 resize at 1440x1152 with VMR9

Or FFDShow Lanczos resize 1776x1728 with VMR9.

FFDShow Lanczos resize at 1767x1728 is definitely my preference.

1776 may sound like a strange resolution, but it is actually my horizontal desktop resolution and 1728 is 3x PAL vertical DVD res.
As fare as I know, FFDShows Lanczos resize is only buggy with non even multiple vertical resolutions so I am using a clean 3x multiple for that.
Even if horizontal resize is buggy at non even multiples, I still prefer 1776 over a clean 2x DVD res of 1440. This could be because 1776 avoids scaling by my Radeon or because higher resize gives a sharper image with VMR9. It’s probably a bit of both.
With a bit more speed out of resize I can go to 1776x2000. That will allow a nice halving of vertical resolution by my Radeon to get back to my vertical desktop res of 1000.

This resize thing can be very tricky to get perfect.
It pays to try out all the possibilities, even if they seem unusual or unlikely.



Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Has anyone else observed this?

Yep, I’ve noticed it as well, but only with the last two SSE enhanced builds. I actually observed a BIG change in hue when the chroma control is cranked. Also, there is kinda a dead spot in the effect at certain luma settings. If I set luma about .5 and crank chroma toward max, most of the red drops out. If I then crank luma, there is a range that it pops back in. Strange. (as well as the big green stripe along the left side at exaggerated settings) Perhaps I’ve got something mis-configured (color-space?) (filter-properties?) but I’ve gone back to the 20030927 build for now as it doesn’t exhibit this behavior and I only saw modest performance gains on my system. Not enough to allow denoise-3D/ resize/sharpen/VMR9 so I've returned to denoise-3D/dscaler sharpen. For now anyway.

Dave

Well, Im still using Owens Denoise3D setting on my lowely AMD 1800 and movies look good to me. WIll keep those settings untill I hear on here otherwise.

Question: I have that first setting in ffdshow set at any or all. Should I set it to YV12 ? Im using Theater Tek. I heard on here that TT uses YV12 at default anyway.

Wow, glad I decided to check this thread again...I can now use denoise3d and then resize @ 1920x1080. Can even do it using VMR9, though I still have a tearing problem due to my 9600 non-pro card. Very nice work, thanks! Tried some of the higher rez settings Owen mentioned, but I don't quite have enough juice for those. ;)

I have also noticed a change in Chroma levels or even a green screen when adjusting Denoise3d settings.
At the settings I use L0.5, C1.0, T5.0 HQ there is only a minor drop in color compared with Denoise3d disabled.
It is easily corrected with color level adjustment. No big problem.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Again... where do you specify the colorspace output type? The only thing I've screwedwith beyond resize, sharpen, and denoise is to make sure that I have RAW checked.

Owen,

What filter do you use to correct the color, and what adjustment do you make?

Also, do you see the band on the right side of the screen that I see? I assume not, since you didn't mention it.

Final question (unrelated), what does the "parameter" adjustment in the lanczos sharpen do? I have it set at "default" but I have seen others set it to "10".

madpoet,

You select the color output space in ffdshow by going to the very bottom of the filter list. You will see something called "Output". When selected, you will see all of the checkboxes for the various options.

Thanks,
Phil

Again... where do you specify the colorspace output type? The only thing I've screwedwith beyond resize, sharpen, and denoise is to make sure that I have RAW checked.

Should be the option under 'Miscellaneous' on the left pane, I can't see what it says becuz im using windowblinds. :)

Ahhhh... ok. Thank you. DO we have a consensus on what colorspace to output?

So for the YV12 output setting in FFDShow, do you need to check the box next to Overlay, or can you just leave that box unchecked and then uncheck all the boxes next to all the other colourspaces other than YV12?

I'm worried that if you have to check the Overlay box, then I'll need to adjust all the colour controls there in to match what I had before I checked it. If I only need to ensure YV12 is checked and that's it, then I know it's not affecting colour overall and only for using YV12.

Madpoet,

Look down the bottom of the FFDShow control page for “Output”
It’s just above the “About” menu item.


Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Ever since I started using resize, all animated menus in a DVD are horribly choppy. This is any dvd, any menu. The actual movie plays fine though. Anyone else run into this?

Yep... CGI openings crush my PC. The opening to Big Fish for example. It also holds true for heavy CGI scenes within a movie.

Common and normal.

YV12 output seems the best option.

You will need to UNCHECK everything EXCEPT YV12.

Leave Overlay Mixer UNCHEKED

Looks like I was a bit slow with my last reply, madpoet. :D

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Originally posted by Owen
YV12 output seems the best option.

You will need to UNCHECK everything EXCEPT YV12.

Leave Overlay Mixer UNCHEKED



What about the Output checkbox which is the main header on that section, does the checkbox need to be on for that, or leave that unchecked and only check the YV12 box?

NTSC intro scenes can be 30fps video.
The main movie is 24fps.
Those 6 extra frames per second can be quite an extra load.

PAL users should not get that problem.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

It's ok, thanks ;). Now I know just where to look. If I'm not too busy drooling over my X800XT in UT2004 tonight.

Originally posted by pmd918
Owen,

What filter do you use to correct the color, and what adjustment do you make?

Also, do you see the band on the right side of the screen that I see? I assume not, since you didn't mention it.

Final question (unrelated), what does the "parameter" adjustment in the lanczos sharpen do? I have it set at "default" but I have seen others set it to "10".

Thanks,
Phil

I just use Zoom Players color control.

No I dont have the band.

I have never seen an explanationof the Parameter control.
It does not seem to make much differance.


Regards,

Owen

____________________________
The FFDShow resize-sharpen dude.

Originally posted by Owen
I have also noticed a change in Chroma levels or even a green screen when adjusting Denoise3d settings.
At the settings I use L0.5, C1.0, T5.0 HQ there is only a minor drop in color compared with Denoise3d disabled.
It is easily corrected with color level adjustment. No big problem.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

I guess my point is is that the older versions don't exhibit this behavior. I'm not sure when the 'bug?' crept in but both the latest versions seem to have it. For me, it just doesn't behave like it should with dead spots and quantum leaps in the controls. Without adjusting R,G,B levels independently, I'm not sure you can correct for it. I couldn't with just saturation. Saturation would bring back the deficient reds but then green was too intense. If others are experiencing it, and the older versions behave properly, perhaps it can be identified and fixed. A denoise algorithm shouldn’t affect the white point or color balance, IMO.

Dave

I got rid of my horizontal tearing problem with VMR9 and 9600 non-pro. I downgrade Lanzcos resize from 1440x960 to 720x480. It doesn't tear anymore. However, at least on a CRT monitor I don't think it looks as good either.

I don't really understand AR well enough. My intention is eventually to drive an 800x600 projector. My CRT desktop is currently set to 800x600 for testing purposes and ffdshow resize is (was) set to 1440x728. I'm not yet a powerstrip user. Do I need to set ffdshow resize to 1440x728, establish a 1440x728 resolution in powerstrip, and run it in a 480p timing?

Thanks

Sorry to ask such a stupid question, but is the SSE2 version only for 64 bit CPU's? I tried it on my Athalon Barton 2500+ (OC'ed to 3200+) and it didn't work. Just wanted to make sure...\

I downloaded the non SSE2 version, and it worked fine. Still having the problem with VMR9 by the way. Which should I be using: windowed or windowless?

Thanks for your help so far! :)

Dlarsen,

Don’t forget that Denoise3d has just undergone some serious recoding.
I’m sure this bug will be quickly squashed.
Just put it down to teething troubles. :D


Regards,

Owen

____________________________
The FFDShow resize-sharpen dude.

Originally posted by basement
I got rid of my horizontal tearing problem with VMR9 and 9600 non-pro. I downgrade Lanzcos resize from 1440x960 to 720x480. It doesn't tear anymore. However, at least on a CRT monitor I don't think it looks as good either.

I don't really understand AR well enough. My intention is eventually to drive an 800x600 projector. My CRT desktop is currently set to 800x600 for testing purposes and ffdshow resize is (was) set to 1440x728. I'm not yet a powerstrip user. Do I need to set ffdshow resize to 1440x728, establish a 1440x728 resolution in powerstrip, and run it in a 480p timing?

Thanks

I suggest that you use Overlay for output to allow you to use a higher resize.
VMR9 looks very blurry at 720x480. Even at 1440x960 it is still to soft for me.

Radeon overlay works very nicely and should work much better with your hardware limitations.

Powerstrip only sets your desktop resolution. Your video card will always scale FFDShow output to your desktop resolution for output to your display.


Regards,

Owen

____________________________
The FFDShow resize-sharpen dude.

Are any of you that are evaluating video, different input and output formats of ffdshow, and overlay mode adjustments, doing this while actually watching a scope connected to one of the outputs of the HTPC?

I have been, and there are some interesting surprises with levels, crushing and other anomalies. Since it was just mentioned, if you're using YUV raw input to ffdshow and YV12 only output, and overlay, in order to maintain the right output levels (100 IRE white should = .7V p-p, and 0 IRE should equal 0 volts (normally), you should engage the ffdshow Levels option (before resize) with the input white tab set at 235, and the output black tab set at 16, full range checked, everything else unchecked. Gamma can be tuned to preference, but mine is usually about 1.18.

If you don't do this, you will be crushing whites and blacks. It's very interesting watching how some controls affect the video when your eye is not the only thing doing the interpreting.

There are also strange things that happen wrt both white and black levels when you select some other options in ffdshow, your player, or your video card's configuration panels. For example, if I have ffdshow set as described and then access the NVidia control panel, the white level raises a bit and the black level lowers a bit. The result is some mildly crushed whites, and blacker than black references are also crushed. If you then go back to ffdshow, these levels correct themselves.

The way I worked around this was in ZP Pro, activating the Overlay levels controls, and querying the overlay. I found that under ffdshow control the contrast level read 9000, but under NVidia's control settings it read 10000. Contrast levels on overlay affect both the black and white levels (black goes down when white goes up). By setting 9000 as the default in ZP Pro, the level stays the same.

There's a number of little things that change when you don't expect it, and can have quite an effect on the way you perceive the picture and conclusions you might reach from time to time. Just knowing that some contrast adjustments affect both white and black levels, while others (e.g. those on the projector or monitor) typically adjust only white level.

Also, many already know this I'm sure, but the Overlay adjustments are different from Desktop adjustments. On the NVidia drivers you can tweak them individually to normalize. Without this normalization color temperature alignments with software such as Colorfacts which preferentially use the Desktop for calibration will produce a standard for which your overlay DVD playback is not correct.

On a slightly different subject, does anyone know why NVidia drivers (59.64) default to a color saturation level of 114%?

--Bill

OK - I give up. Latest FFDShow version i can find is May 04.

Where do i find the latest 20040610c_SSE2 ?

Cheers
Bill

Bill,

Yep... have my scope sampling a line of video most of the time but am only using VMR9 since I am profiling the video which is not possible (in my setup anyway) with overlay. I use Optical/Spyder combo to build the profile.

I generally use "Picture Properties" to crank in a luma (labled luminance) offset of -16 and then use "Levels" with an input range of 0 to 219 and an output of 0 to 255 - it is a little more convenient set but both are accurate. Many DVDs require the gamma that you mentioned. I use from 1.0 to 1.5 with different sources. This gamma adjustment is mostly due to my projector having low output.

This scaling works great with one exception - it loses every 7th sample, or a total of 31 levels. This is unavoidable and has been an issue that has caused me to scrutinize many greyscales trying to see if I can spot the missing levels. So far I can't see any difference at all even if I switch back and forth.

I also have used the projector controls to do the same thing as ffdshow and the results are good. This technique gets away from the lost levels problem but doing the gain scaling in analog circuitry. It has two disadvantages in that the profile is a little hard to get everything right without a little tinkering and that any signal from the PC that is not video will have a larger range and cause the whitepoint to saturate.

I have been pleased with the results from letting ffdshow do all the work and am making a little setup DVD, tonight, that has levels that fit in with how Optical works so tweaks will be easy.

Have fun tinkering!

Charlie

Originally posted by Bill Green
OK - I give up. Latest FFDShow version i can find is May 04.

Where do i find the latest 20040610c_SSE2 ?

Cheers
Bill

Try here. (http://mitglied.lycos.de/ieggei2/ffdshow/)
_____
Axel

Thanks Axel - will try it out tonight.

Bill

I have been, and there are some interesting surprises with levels, crushing and other anomalies. Since it was just mentioned, if you're using YUV raw input to ffdshow and YV12 only output, and overlay, in order to maintain the right output levels (100 IRE white should = .7V p-p, and 0 IRE should equal 0 volts (normally), you should engage the ffdshow Levels option (before resize) with the input white tab set at 235, and the output black tab set at 16, full range checked, everything else unchecked. Gamma can be tuned to preference, but mine is usually about 1.18.

Hi Bill,

I am no expert in these things but very willing to learn ... :)

Could you tell me if this adjustment is also necessary in VMR9 and with DVI output ?

Maybe thats the right place to ask another question ( I posted this in the ffdshow development faq already and Andy answered but I would just like to get a more detailed explanation for this ):

On my setup ( FX5700 GPU, VMR9, NVIDIA 6x.xx drivers) different output color spaces ( YUV2/ YV12 versus RGB32 ) require totally different brightness/contrast settings on my projector when calibrating it with test patterns. For RGB32 I have to increase brightness and decrease contrast quite a bit.

I was just wondering why there is so much difference between colorspaces. Some people seem to prefer RGB32 to YUV/YV12 but nobody seems to mention the difference in brightness and contrast associated with it.

Thanks,
Alex

Hi there
Regarding Denoise3D: I`m using the latest SSE2 optimized build which exhibits a major color bug (Not sure if previous versions also have this bug)! My current configuration is Elecard->Levels->(Denoise3d 0.5 1.0 5.0 HQ)-> either Avisynth Lanczos4resize to 720p or ffdshow Lanczos to 1440*1152->forced YV12 output->VMR9->DVI->Sharp Z10000 projector. This color bug has been verified with WINDVD 5,6 deoders, Overlay, YUV2 output,... all with the same result: Switching Denoise3D on/off results in different colors (clearly a chroma bug as grayscale is not touched) It can bee seen by naked eye and I have verified it with screenshots (checked RGB values with Photoshop). Just play a video (I used the demo sequences of DVE) and switch Denoise3D on/off. Depending on the scene you should be able to clearly see a significant change in colors. That makes Denoise3D IMO unusable right now, as it comes with the price of wrong colors.

Another bug to report: The Lanczos resize filter in ffdshow still has the green bug if you set chroma sharpening >0.6 Below that value grayscale rendering is ok (with direct multiples of source resolution). Adding chroma sharpening above this value results in the well known green bug again... So avoid using chroma sharpening...

Originally posted by Owen
I suggest that you use Overlay for output to allow you to use a higher resize.
VMR9 looks very blurry at 720x480. Even at 1440x960 it is still to soft for me.

Radeon overlay works very nicely and should work much better with your hardware limitations.

Powerstrip only sets your desktop resolution. Your video card will always scale FFDShow output to your desktop resolution for output to your display.


Regards,

Owen

____________________________
The FFDShow resize-sharpen dude.

Using powerstrip I increased my desktop resolution to 1440x960 and with overlay it looks fantastic. However, I'm still having trouble with reliably launching zp with overlay enabled. 5 times out of ten, zp hangs at start up. It switches to a black background then nothing. I have to go into task manager to kill it. Running without ffdshow it always comes up. I think I'll try re-installing everything again.

Thanks for your comments and insight,

Peter

Originally posted by Projectorguy1
Sorry to ask such a stupid question, but is the SSE2 version only for 64 bit CPU's? I tried it on my Athalon Barton 2500+ (OC'ed to 3200+) and it didn't work. Just wanted to make sure...\

I downloaded the non SSE2 version, and it worked fine. Still having the problem with VMR9 by the way. Which should I be using: windowed or windowless?

Thanks for your help so far! :)

I guess it really was an off the scale dumb question, but I havent been able to find the answer and this is a REALLY long thread. Is the SSE2 a more MMX optimized version I should be able to run on an Athalon system?

the sse2 version can't be used in amd cpu gears.

How does one set the order in which the settings in FFDSHOW are run?

Also what is the best order when using resize and denoise?

basement, weird that you should mention that problem, but I get that sometimes, and it's quite random too.

Projectorguy1, if you have an A64 you should be able to run the SSE2 version, but if yours is an Athlon(not Athalon btw) XP then no you're outta luck

chen I do believe AMD A64/Opteron systems will be able to run the SSE2 builds since they support SSE2, no?

rgreenpc, once you have a particular setting highlighted, either click on the up/down arrows to the left of the setting name, or hold down your shift key while using your up/down arrow keys to move them up/down.

Goi -

Thanks about 6 seconds after I posted I noticed the little arrows - LOL

Is there a preferred or best order - like denoise then sharpen then resize or.... ?

i just made a resize as follow:
ndvd 3.0+ffdshow(no ndvdvpp)
ffdshow set as: 1920X1080,lanczos,luma sharpen 0.88,blurr & nr Gradual denoise 20,xsharpen 20,150,
the order:blurr&nr>sharpen>resize&aspect
it is very film like.
btw,i don't think andy's ffdshow 20040607a ease my barton2500 cpu usage a lot ,it just 3-5% in my pc.the above tweak load 90% of my cpu.
the zoomplayer+ffdshow+windvd6 with the above setting works fine too,images become a little sharper,but not so film like,and color is less richer than nvdvd 3.0 setting.
and,i prefer the xsharpen in the above settings to the unsharp mask,because the latter cause much grains.
if put the xsharpen behind resizer,images very fine and clean,but cpu load 100%,and get shutter.
unsharp mask put behind resizer is ok,but not as good as the above setting.

I tried changing the setting from ALL to YV12 as suggested on here in that first menu of ffdshow. I guess thats where your suppose to make this adjustment. Well, I can't start Theater Tek now. I had to go back and reset that setting back to ALL. Cant remember what its called exactly. Its what you've all been talking about the last week or so. Its the setting you have to set or ffdshow will lock up TT if you dont set it to ALL. RAW video I think its called.

rgreenpc, if you've been reading the past couple pages of this thread, Owen's current recommendation is denoise3d followed by Lanczos resize with luma sharpening. Theoretically resize should be done first if you want max image quality, but there's no current CPU that can do resize followed by denoise3d fast enough.

I tried changing the setting from ALL to YV12 as suggested on here in that first menu of ffdshow. I guess thats where your suppose to make this adjustment. Well, I can't start Theater Tek now. I had to go back and reset that setting back to ALL. Cant remember what its called exactly. Its what you've all been talking about the last week or so. Its the setting you have to set or ffdshow will lock up TT if you dont set it to ALL. RAW video I think its called.

styleinlp,

you have to leave that setting at "all supported". Output Colourspace is changed at the very bottom of the ffdshow filter tree. Look for "Output" ... there you change your output colorspace !

Bye,
midiboy

Is the Information & Statistics panel in ZoomPlayer accurate when used with ffdshow? I have had it up for a while now under a stress test (not of ffdsshow, but of my P4 3.0c, now overclocked to 3.833 mHz, apparantly rock stable since 3.9 mHz gave me one crash after around five hours of stresstesting).

Right now I'm about halfway in Return of the King...after having played the EE of Fellowship and Two Towers. All this with the slightly extreme settings I have mentioned before (dScaler sharpness, Levels (to correct that white/black crushing - does it happen on Radeons as well btw?), deNoise3D and finally resize to 1920x1728. If the Information & Statistics is correct it hasn't dropped one frame all this time! CPU use is jumpting between 88 and 99, most of the time it's on 93.

Thanks Midiboy. Will do that tonight :)

Charlie,
It sounds like you're using a slightly different approach than I am, by sampling in real-time on program content. I'm sure there are lots of interesting discoveries in doing that, but I'm always referring back to a particular pattern produced by Avia Pro. It's just a 100 IRE white block about 1/3 the size of the screen, over a black background, and is stationary. So you have a steady state reference that quickly shows any kind of fluctuation in the 0 to 100 IRE range while adjustments are made.

[IRE actually is supposed to refer to an absolute voltage level from 0 to 1v with 1v representing full screen brightness, but in current use it's a video level reference in which 100 IRE is max white, which corresponds electrically to .7v on each R,G,B output]

What was especially a surprise to me was how many of the adjustments affect both white and black level simultaneously rather than just one or the other. By comparison, the contrast control of a pj or monitor affects only white level, and brightness affects only black level. I've attached a scope picture of my reference pattern. It's amazingly easy with minor adjustments to crush the two levels right below max white without noticing it (initially), and the blacker than black reference gets skewed or lost easily as well. This shot is .1v per division vertically.

midiboy,
Calibration is needed in any format and any type of output if you expect to achieve the best results. Overlay and VMR9 are just different ways to write to the video card, so with regards to output voltage levels they are calibrated in the same manner.

Real DVI output (not the analog lines that are also accessible on some DVI connectors) can't be directly read with an analog scope on the outputs, but it too is subject to calibration adjustments within ffdshow in exactly the same manner. The difference is we can't assess it in terms of an analog voltage.

On the colorspace question, it is not unusual that different colorspaces require different settings, since the actual range from black to white may contain a different number of levels. When ffdshow is doing colorspace conversion, it would be mandatory to make level adjustments. I wouldn't think the correct place to make these adjustments is on the projector, though. Normally, the projector (or any display device) should be fairly normalized to produce full white output at .7v p-p, and full black at 0v (or slightly higher if you're using the 7.5 IRE black level reference).

The colorspace calibration of the output device (video card) should be calibrated so it correctly produces its full range with the same voltage levels, which gives you maximum electrical accuracy between the physical devices. ffdshow would then be used to calibrate black and white levels to correspond to the range required by the display card and downstream.

Obviously, you can tweak and adjust anywhere in the chain, but the results will not always be optimum. For example, if your white level was too loo low going into the overlay, which would make the voltage low on the output, and then you raise contrast on your projector to make up for it, you're creating two adverse situations. One, you're using a lower digital range (effective number of bits) in the transfer from ffdshow to video, then you're amplifying that video card output at the projector with the contrast control. So the two adverse issues are less black to white 'steps' making the steps coarser, and then amplifying the analog signal at the projector which can increase noise and anomalies like ringing from the cabling. So while your black to white range is restored, the overall picture can suffer significantly.

Conversely, if the white level was a little too high going to the overlay and you reduced contrast at the projector you have two different possible scenarios. One, there would be white crushing occurring at the overlay which could not be recovered by reducing contrast later. Two, you may exceed the video card's voltage output capabilities causing crushing or non-linear distortion, which again could not be corrected by lowering contrast at the projector.

I would even go a step further by saying that many such individual 'preferences' in colorspace, levels and other processing are determined in a mis-calibrated state, adding significant bias to the results. If you don't know for sure where your levels are and whether or not you're crushing whites or blacks (even just a little), you may be missing a great deal of picture quality.

--Bill

bblue, are you sure you meant that the input tab should be 235? Maybe you meant 35? On 235 the picture on my monitor is as red as the blood in a John Woo movie... :cool:

bblue, I do know that people that use Theater Tek set their video settings in TT at certain levels prescribed by experts on the TT forum.
Also levels are effected by my Extron unit. I have my little recessed screw driver adjustment cranked 100%. That adjustment is called Level.

Are you saying that this isn't enough?

Obviously, you can tweak and adjust anywhere in the chain, but the results will not always be optimum. For example, if your white level was too loo low going into the overlay, which would make the voltage low on the output, and then you raise contrast on your projector to make up for it, you're creating two adverse situations. One, you're using a lower digital range (effective number of bits) in the transfer from ffdshow to video, then you're amplifying that video card output at the projector with the contrast control. So the two adverse issues are less black to white 'steps' making the steps coarser, and then amplifying the analog signal at the projector which can increase noise and anomalies like ringing from the cabling. So while your black to white range is restored, the overall picture can suffer significantly.

Conversely, if the white level was a little too high going to the overlay and you reduced contrast at the projector you have two different possible scenarios. One, there would be white crushing occurring at the overlay which could not be recovered by reducing contrast later. Two, you may exceed the video card's voltage output capabilities causing crushing or non-linear distortion, which again could not be corrected by lowering contrast at the projector.

I would even go a step further by saying that many such individual 'preferences' in colorspace, levels and other processing are determined in a mis-calibrated state, adding significant bias to the results. If you don't know for sure where your levels are and whether or not you're crushing whites or blacks (even just a little), you may be missing a great deal of picture quality.



Hi Bill,

thanks for your detailed and very informative answer ! You certainly seem to know a lot about this !

Do you mind if I ask you to maybe give some practical tipps on how one should go about calibrating if one does not have a scope and/or is using DVI connection to the projector ( and I think I am really staying in the digital domain when going from my Geforce 5700 to the Sim2 HT300+ ) ?

Until now I just "assumed" that levels from the videocard are sort of neutral and did all my adjustments on the projector. The fact that different colorspaces required different calibration settings on my projector got me thinking though.

So if one does not have a scope ... how should one decide upon which end of the chain to use for calibration ?

I hope this has not been answered before in this thread. I have not read through all of the 49 pages. :)

Thanks,
midiboy

mastiff,
i'm saying that the WHITE tab of the input bar which is normally at 255 should be set to 235. The BLACK tab of the output bar which is normally at 0 should be set to 16.

No matter what you do there, there should be no variance in color. If there is, the grayscale of your monitor or pj may be off. These are overall video levels if you have 'Modify only luminance' UNchecked, and 'Full Range' Checked.

Stylinlp,
I would assume (you know what they say about assumptions, though) that the number adjustments made by the TT experts are based on measured levels -- I seem to recall reading that somewhere. When all pieces of your setup are known commodities you can recommend numbers that should produce predictable responses. However, other adjustments that might be made by the user could have an effect on all that. You can't be sure without measuring the output.

On your Extron, I don't know what the 100% range means. On a typical buffer or amplifier that has adjustments you would want them set so that what comes in is exactly what goes out. If you actually have to increase voltage (gain) chances are you are making up for a mis-calibration on the HTPC video output, or on the projector.

--Bill

Aha. That explains it. And believe me, the grayscale, the sharpness, the everything is off on that monitor! It's an old 15" computer monitor, I haven't hooked this new HTPC up to the Barco yet! ;)

midiboy,
You can't do these adjustments in the analog world without a scope. There is just no other way to measure.

If you are connecting to your projector to a DVI input, it's an all-digital path. If you are coming out of the HTPC DVI jack with a VGA adaptor and then in to a VGA or RGB input, it's not digital. I don't really know what the procedure would be to calibrate your video card or ffdshow controls, unless the card drivers have that information.

Most of the time, the card drivers are pretty close given that they receive a conforming signal via Overlay or VMR9. But since we have so much control over that signal with ffdshow, it's quite possible to be way off. The conversions between colorspaces should be a predictable and known quantity, so if you knew what they were you could come very close to right on with just Levels and Gamma. But there are unexpected surprises that you could easily miss by not seeing a scoped output.

I've never been able to get RGB32 output working on my system here, there's always a filter connection error in ZP Pro that comes up on play when that is the only output option. If that was working I could scope it and see what is needed.

--Bill

Whatever happened to seeing what you liked and adjusting it from that? Sigh... :D

Bill,

I hope you don't mind boring details... so here goes.:)
I'm sure there are lots of interesting discoveries in doing that, but I'm always referring back to a particular pattern produced by Avia Pro. It's just a 100 IRE white block about 1/3 the size of the screen, over a black background, and is stationary. So you have a steady state reference that quickly shows any kind of fluctuation in the 0 to 100 IRE range while adjustments are made.
I am not sure which Avia screen you are referring to but many have level 16 for black and level 235 for white in the US version I looked at. 100IRE would be level 255 (0.714 volt) in 480i. DVDs in the US are like this but the rest of the world DVDs use level 0 as black and level 219 as white.
[IRE actually is supposed to refer to an absolute voltage level from 0 to 1v with 1v representing full screen brightness, but in current use it's a video level reference in which 100 IRE is max white, which corresponds electrically to .7v on each R,G,B output]
In 480i (DVDs) 100IRE is 0.714 volt but in the newer EBU N10 standard (component video, HDTV, 576i) maximum white is 0.700 volt and there is no setup. I don't know if IRE nomenclature is appropriate for EBU N10.

What was especially a surprise to me was how many of the adjustments affect both white and black level simultaneously rather than just one or the other. By comparison, the contrast control of a pj or monitor affects only white level, and brightness affects only black level.
This is why a CRT calibration should start with setting the black level and the white level along with setting the white color temperature to D65. Avia and Precal are good for this. What I do (VMR9) is set the projector contrast control to a reasonable level and then use the CRT gains and bias's to get my blackpoint and whitepoint (at D65) where I want them. Then I run Optical with the Spyder sensor and produce an icm profile and do a validation run too. This gives graphs of each tube's gain curves and shows the deviation from desired gamma. Each tube should has a gamma close to 2.5 and if there is much deviation I adjust the contrast control up or down as appropriate and do the calibration again and again and... untill the projectors RGB outputs are fairly close to gamma 2.5. It is easy though. After the CRTs are about gamma 2.5 the icm profile will have the least amount of correcting to do.
I've attached a scope picture of my reference pattern. It's amazingly easy with minor adjustments to crush the two levels right below max white without noticing it (initially), and the blacker than black reference gets skewed or lost easily as well. This shot is .1v per division vertically.
Nice picture! I wish I had a camera.;) Your scope picture shows that setup is still there and that the level 235 needles are at about 251. Does this include some luminance gain from ffdshow or does overlay scale the output?

I scanned two of my recent Avia scope printouts so you can see how VMR9 output looks but I don't know if they will be legible. VMR9 is right on as far as luminance goes but UV accuracy is unknown by me.

Charlie

Bill,

Lost the attchment somehow. Will try again

Charlie

Originally posted by Charles Black
I am not sure which Avia screen you are referring to but many have level 16 for black and level 235 for white in the US version I looked at. 100IRE would be level 255 (0.714 volt) in 480i. DVDs in the US are like this but the rest of the world DVDs use level 0 as black and level 219 as white.


For both NTSC and PAL DVDs, the digital levels range between 16 and 235 (actually there are slightly different ranges for the luminance and chroma signals, but they are the same for both standards).

It is only when the player begins the analog conversion that voltage levels are differentiated and offset from zero or not according to the desired output standard. The digital range is exactly the same, only the analog is altered to meet local expectation. This is why Japanese discs play fine in the US (aside from being Region 2), even though Japanese NTSC-J is based on 0 IRE.

If you are only looking at the analog outputs, it would appear as you suggest but the digital range prior to analog conversion is the same.

John,

--------------------------------------------------------------------------------
Originally posted by Charles Black
I am not sure which Avia screen you are referring to but many have level 16 for black and level 235 for white in the US version I looked at. 100IRE would be level 255 (0.714 volt) in 480i. DVDs in the US are like this but the rest of the world DVDs use level 0 as black and level 219 as white.
--------------------------------------------------------------------------------

For both NTSC and PAL DVDs, the digital levels range between 16 and 235 (actually there are slightly different ranges for the luminance and chroma signals, but they are the same for both standards).
Thanks for catching that! I am used to thinking of levels as mv and should have said that in places where North America style setup is not used level 16 (reference black) is set to 0mv and level 235 (reference white) is set to about 614mv in analog outputs.

This is assuming that the DAC is set to output 714mv full scale, which might not be the case at all in invididual video cards. My nVidia TI500 card puts out 710mv into my projectors RGB inputs at level 255 but small changes in input impedance will cause this value to change a little. It would seem logical that the best design choice for level 255 output would be 700mv in light of the 576i standard so there may be a small range (say 700mv to 714mv??) of full scale voltage differences between video board RGB outputs.

Let me know if I have still got this screwed up.:)

Charlie

Originally posted by bblue
i'm saying that the WHITE tab of the input bar which is normally at 255 should be set to 235. The BLACK tab of the output bar which is normally at 0 should be set to 16.
Interesting, if the level range is 16 to 235 on the DVD, why do you set the input white tab to 235 while setting the output black tab at 16? Can you explain the reasoning behind not setting just the input tabs, to expand the range from what's on the DVD to the capabilities of the display device?

If I set the output black tab to 16, black levels just get worse (like increasing brightness), so my displaying is making use of those 0-16 values. Why is a setting 0, which increases the number of grey scale levels, bad?

I use the Avia cross grey scale ramp and DVE shallow ramps to adjust this, making sure that it's linear all the way across, from pure black to pure white, with no blocks of solid white or black. Is there a better adjustment to use on Avia or DVE?

Originally posted by TheLion
Hi there
Regarding Denoise3D: I`m using the latest SSE2 optimized build which exhibits a major color bug (Not sure if previous versions also have this bug)! My current configuration is Elecard->Levels->(Denoise3d 0.5 1.0 5.0 HQ)-> either Avisynth Lanczos4resize to 720p or ffdshow Lanczos to 1440*1152->forced YV12 output->VMR9->DVI->Sharp Z10000 projector. This color bug has been verified with WINDVD 5,6 deoders, Overlay, YUV2 output,... all with the same result: Switching Denoise3D on/off results in different colors (clearly a chroma bug as grayscale is not touched) It can bee seen by naked eye and I have verified it with screenshots (checked RGB values with Photoshop). Just play a video (I used the demo sequences of DVE) and switch Denoise3D on/off. Depending on the scene you should be able to clearly see a significant change in colors. That makes Denoise3D IMO unusable right now, as it comes with the price of wrong colors.

Another bug to report: The Lanczos resize filter in ffdshow still has the green bug if you set chroma sharpening >0.6 Below that value grayscale rendering is ok (with direct multiples of source resolution). Adding chroma sharpening above this value results in the well known green bug again... So avoid using chroma sharpening...

oki i will take a look tomorrow and see if i can find and smack this bugs thx for the info

Your scope picture shows that setup is still there and that the level 235 needles are at about 251. Does this include some luminance gain from ffdshow or does overlay scale the output?

I scanned two of my recent Avia scope printouts so you can see how VMR9 output looks but I don't know if they will be legible.Charlie,
How did you determine that the 235 needle was at 251?

I don't think overlay is doing any scaling of its own, but I had adjusted the overlay value of contrast from 10000 to 9000. Without doing this, there were no needles visible on the high end -- they were completely crushed, and the blacker-than-black was crushed as well, though not completely. I simply lowered the overlay contrast setting until there was no distortion of the white level steps, and blacker than black was just starting to compress.

On your picture, I can make out the shapes and all, but none of the writing is legible so I'm not sure what the notes are communicating.

It was interesting to note that my maximum white level output from the video card seems to be just slightly under 7v, unless I crush hard into the whites.

I'm not quite sure what gray scale calibration of the projector has to do with this directly, but for that I use Colorfacts for the initial tracking and all. Ultimately across several sources, master black level (brightness) and white level (contrast) are adjusted by eye per source. The gamma results that Colorfacts produces/plots don't seem to agree with what I see on source, or what adjustments are made to produce the final results -- for that it's pretty much by eye.

nm88,
The need for the Level adjustment that I'm doing seems to be directly related in ffdshow to how the input and output colorspaces are configured. In my setup, input is set to raw-all YUV, and output is set to YV12 only. Other combinations don't produce the same adjustment needs. Once at that configuration, I first adjusted the Level settings by eye -- what combination produced results that were the most standard to my other HD sources (D-VHS, 3 different HD SAT systems, OTA HD and Momitsu DVD (modified)), and after determining that I looked at the scope to see how it stacked up. I was a little off on the high setting (242 vs 235) but 16 was right on (by watching the histogram).

I can't say why you adjust one setting on input and the other on output, but that is what seems to produce the right results. Maybe someone else can shed some light on this?

--Bill

awwwwwwwwwwwww my brain sooooo sore..............

haha

Since there is discussion on levels, I thought I'd chime in. During my recent visit to NY home entertainment expo and having Joe Kane look at my HTPC connected to his Samsung 720p DLP projector, he did not mention any issues with levels. I had some issues with ATI in the past and found that cutting at 16-235 helped. My Nvidia setup that Joe Kane looke at was:

XP Pro, DX 9.0b, Nvidia Forceware multimedia 3.0 Beta, in software mode, VMR7 connecting to overlay, Avisynth (v 2.5.5) Lanczos resize to 960x540, FFDShow (March 25th build) resize Bicubic to 1920x1080, Nvidia Quadro FX 1100 with 56.72 official drivers, forced AF 8x.

I still see no issues with levels with DX 9.0c, 61.12 drivers, VMR7 (not connecting to overlay) and AA forced to 4x (AF still 8x, as well as other settings and setup).

Lets get this clear, as I can see confustion in the questions. We're talking about setting INPUT levels 16 and 235, and OUTPUT levels 0 and 255 right?

So we take the range of levels in the DVD source, and then expand to the full range for the display device?

[QUOTE]Originally posted by vpopovic
[B]Since there is discussion on levels, I thought I'd chime in. During my recent visit to NY home entertainment expo and having Joe Kane look at my HTPC connected to his Samsung 720p DLP projector, he did not mention any issues with levels. I had some issues with ATI in the past and found that cutting at 16-235 helped. My Nvidia setup that Joe Kane looke at was:

How do i make FFDShow active ONLY when SD PAL is playing and not HD?

Currently in Zoomplayer i have it set after the Elecard Codec in the customised media for MPEG 2 and works great, but tries to work with 1080i HD, and well my cray super computer is yet to arrive :)

I tried under "image settings" in FFDShow setting "preset auto load" conditions on the default config to 720 x 576 but it still loads for HD.

Do i need to perhaps leave the default setting with NO FFDShow options check and then make a second config that enacts with 720 x 576 PAL source?

hmmmm might try that :)

Originally posted by gazzagazza
Lets get this clear, as I can see confustion in the questions. We're talking about setting INPUT levels 16 and 235, and OUTPUT levels 0 and 255 right?

So we take the range of levels in the DVD source, and then expand to the full range for the display device?



I think I had it the other way around. Input full, output clipped at 16-235. At that point RGB32 output from FFDShow also looked better that YV12. However, this was with FireGL drivers, not Catalyst. If I remember correctly, the problem was gone using January version of FireGL drivers.

Originally posted by vpopovic
I think I had it the other way around. Input full, output clipped at 16-235.
I'm beginning to think it's very dependent on the output device. After tinkering with it quite a bit with my ATI 9600 card and Pioneer plasma, it's obvious on my setup that any level adjustment -- input or output -- only hurts PQ by crushing whites or blacks respectively.

This is most clear on the shallow grey ramps of DVE (title 12, chapter 15). If I clip the input levels to 16 and 235, the shallow ramps on the top (white) and bottom (black) go from nice smooth ramps (with no ffdshow levels) to solid blocks of white and black, respectively. No other black/white level correction on ffdshow or the plasma will correct this once level clipping is on, that detail simply disappears. PQ while watching any DVD is noticeably worse, particularly with all the black level detail gone.

If I clip the output levels, I just lose grey scale levels on the plasma. I can recalibrate the black and white levels to compensate for the brighter blacks and darker whites, but then the grey ramps consist of fewer 'levels' of grey and look blockier and obviously worse. The plasma is obviously making use of 0-16 and 235-255 so why clip it?

My results may not apply to other display devices, but for plasma owners, I'd suggest doing an actual image comparison before blindly tweaking levels. For reference I'm using WinDVD 6 filters, the latest ffdshow 0610c, Dscaler LumCrom shift, denoise3d, Lanczos 1440x960 resize/sharpen, YUY2 output and VMR9. YV12 output has the same issue.

I'm using dvi out to digital fp.
As i understand it ,since dvd disks are encoded using 16-235 range adjusting levels setting in ffdshow we have to choose what is best between these two.

1. Adjusting input levels to 16-235 and leaving output to 0-255 you expand the dynamic range by mapping 16->0 and 235->255 .
So we have more steps between black and white.
But doing this we loose all below black and above white information.

2.Adjusting the output too to 16-235 or leaving both at 0-255 we have the full dvd disk's signal and also the above white and below black info which we compensate by using our brightness and contast contols.
But in this case we have less steps between black and white.

So we have to decide which is the best.

If my above understanding is correct can someone with the appropriate tools do some testing and inform us?

nm88,
Try it as I described in my original message about this. Input white at 235, output black at 16, and gamma somewhere around 1.2 . With raw video enabled and receiving a YUV type format, and an output to YV12, this is the only combination of settings that show the right output on the scope. That's true on overlay or VMR9. On the NVidia cards (that's all I'm able to test) in the 'analog monitor' section, you can adjust contrast between 9280 and 9980. 9980 is similar to 10000 the default, except it shows 98 and 99 IRE more precisely. They are slightly crushed otherwise. This contrast adjustment affects both white and black levels, so as you bring the setting down towards 9280 the white level comes down slightly, and the black level comes up somewhat. At 9280 the black level is very close to 7.5 IRE and the blacker-than-black needle is just above 0 IRE.

Since you're using VMR9, you could also use the RGB32 output which behaves differently level-wise. On it, your white level is already at max voltage out without adjustments, and black level is at 0 instead of 7.5 IRE. I found it desireable to use the Level adjuster setting Output black to around 16, which brings it up to approx 7.5 IRE and you can move it around between those two points to suit. You can also adjust contrast on the card drivers analog monitor panel down just slightly, and by doing so you will be able to see the 101 IRE whiter-than-white needle if it's present on your test source.

--Bill

It definitely depends on the display device and whether your display works with 16-235, 0-255, or both, and how it internaly processes RGB24 signal. For example, my Samsung RP DLP TV has quite different internal colorspace than RGB24, so I get an extra colorspace conversion on top of what is going on in video card.

However, it also depends on drivers and how they sort through colorspace conversions. I don't think YV12 input to video card is part of WQHL test, so if you are outputing that signal to video card, there might be bugs in the drivers (i.e. it is not standardized how should this conversion be done). Video card needs to convert to RGB32 (your desktop/system colorspace, and while doing that upsample chroma to 4:2:2) and then convert to RGB24 for output to DVI display.

Charlie,
How did you determine that the 235 needle was at 251?I just did a rough estimation by assuming that your RGB output was the the same as mine at .714v for full scale. Then I just took the ratio of your scope level of .700v and multiplied it by the ratio .700/.714*256=251. If your maximum output is different then just put your value in instead of mine.

I don't think overlay is doing any scaling of its own, but I had adjusted the overlay value of contrast from 10000 to 9000. Without doing this, there were no needles visible on the high end -- they were completely crushed, and the blacker-than-black was crushed as well, though not completely. I simply lowered the overlay contrast setting until there was no distortion of the white level steps, and blacker than black was just starting to compress.I don't use overlay anymore and have never looked into any of the adjustments that effect it. Overlays lack of support for icm profiles drove me to VMR9. Why do a two point calibration when you can do an 11 point one? Even so, a general idea would be to try to not have any adjustments to your video except in ffdshow and your projector for test purposes. When I was using overlay I never noticed that it was clipping of had any bad habits and I didn't use any adjustments to it at all.

On your picture, I can make out the shapes and all, but none of the writing is legible so I'm not sure what the notes are communicating.The combination of hard to read original and bad preparation was too much.:) I don't know how to put more than one image per post without putting them on my tiny (5mb) web page. I'l just put them on subsequent posts.

It was interesting to note that my maximum white level output from the video card seems to be just slightly under 7v, unless I crush hard into the whites.Use any RGB 255 255 255 image to get your full scale voltage, then use that to calculate the voltage levels of any RGB levels in your system as needed. Windows uses 255 full white for many of it's screens so white is usually easy to find. I just use Photoshop to make my targets.

I'm not quite sure what gray scale calibration of the projector has to do with this directly, but for that I use Colorfacts for the initial tracking and all. Ultimately across several sources, master black level (brightness) and white level (contrast) are adjusted by eye per source. The gamma results that Colorfacts produces/plots don't seem to agree with what I see on source, or what adjustments are made to produce the final results -- for that it's pretty much by eye.What kind of sensor do you have with Colorfacts? They are a bit pricey for me at $3000 so I had to shoot a lot lower as far as initial investment was concerned.:) Greyscale color temperature tracking is important in limiting those annoying tints that can shift color with changes in image luminance. Also the greyscale luminance levels from your screen are representative of system gamma as well as displaying any luminance video errors. All of these help to reproduce the artist's intent and make our video experience better.;)

Charlie

Does anyone know what new control has been put in to ffdshow with the newest sse2 versions? It says something about control from girder. Can someone explain? Thanks!

@TheLion oki i could reproduce the error sorry for this bug, will be fixed soon. Im taking off the SSE2 version, this bug isnt in the normal MMX2 version.

What have I done? What have I done?!?!? :( I just spent an hour in the home theater flipping back and forth between the new HTPC and the old one. The old one has 1024x768 + dScaler sharpen, the new one has denoise3D, dScaler sharpen and very high max resize, plus levels (according to the discussions here lately). I had problems seing any difference at all! And whats a lot worse: The image seems worse than it was when I had the former HTPC up and running. Now it's slightly changed with a Radeon 9600 instead of the 9800 SE (since the SE has moved to a new home in the new HTPC, so that's what I was watching there), and the latest 4.6 catalyst. Same driver on both. I didn't really spend any time tweaking the image since I was so blown away by the miniscule difference. So any ideas what I have done wrong? I used the same timings in PowerStrip (1024x768 @ 75) and the same overlay settings (color and so on). Have I made a costly mistake? Well, not that costly, I had to get a new computer anyway since my daughter needed a new one. But my hopes were so high, and now I'm about to get drunk and drown my sorrows here! Help!

Don’t panic Mastiff. :D
You just need to do some fine tuning.
For starters, if your old system was working the way you liked it without using the “Levels” filter, what makes you so sure you need to be using it now?
The use of Levels is very system dependant. Some need it and some don’t. To use Level settings that others say works well on their display is a big mistake; your display almost certainly has different requirements.

Establishing the ideal resize for your display will be time consuming, but with a powerful system you have a lot of flexibility.
As you are aware, most people use multiples of DVD resolution when resizing, but multiples or half multiples of display resolution can also work well, or even a combination of both.
For example you could try 1536x1728 for 1.5 times display width by 3 times PAL DVD hight or 1440x1536 for 2 times DVD width by 2 times display hight. The combinations are many.
With a powerful system you can also get the best out of VMR9.
VMR9 used at low resolutions gives an overly soft image, but at high resize settings this problem is greatly reduced allowing you to have the image smoothness and nice colour of VMR9, combined with sharpness normally only available with Overlay. It’s a very impressive combination.

Once you have established a good resize setting you can experiment with sharpening if required. Either resize sharpening which works great at high resize settings and / or other sharpen filters like Dscaler sharpen used before or after resize.
NOTE: Don’t sharpen Chroma in resize as it will only increase noise.

So don’t despair. Great results are there for the taking.
Once you have the new system fully tuned, you will never want to go back to the old system.

Regards,

Owen

___________________________
The FFDShow resize sharpen dude.

Mastiff,
EDIT: my response just crossed with Owen's so his comments are probably better as far as detail of what needs to be done. My comments are on general level.

Differences are what they are (look at Owen's post for optimization). I still have people come over and prefer DXVA image with VMR9 saying it is more "filmlike". Testing with DVE will clearly show you difference between advanced scaling (if done right) and ATI scaling. As far as other filters such as sharpen and denoise, I don't use them. My personal taste is to stay as close to DXVA with better scaling. I get some denoise effect with AA and AF features, but with slight gains in 3D appearance and color rendering. I'd say that IMO good scaling (as posted previously in this thread) combined with AA and AF gives me some 30-50% improvement in PQ. Some people would probably say it is more like 10-20%.

By the way, did you ever hook up HPTCs on your projector, or are you still evaluating the image with PC monitor. If you blow the image to 100" diagonal, it helps appreciate the improvements that you thought were subtle or non-existing.

Thanks, Owen! Does that mean that the three beers and one cider I have had after I wrote that message was wasted? ;) Never mind, I celebrated the swedish victory (Norway isn't in the european cup this year, so I'll have to root for Sweden since I make my money from there) with them! :D

OK, I guess that what I need are a few hours of quality time, setting it all up from the ground. I will also experiment with VMR9, my new computer can do max resize with VMR without skipping a beat. I'll tell you how it comes along, but it may take some time. I'm alone with the dogs right now (no, not divorce, the other option: wife and kids at my in-laws) which means that I work 16 hour days! Anyway, thanks for the encouragement, Owen! :)

This is on the projector, and it's a lot larger than 100" diagonal...the picture's around 8 feet wide!

Mastiff,

Just few questions about your PJ. What is the max resolution your PJ can resolve at this point? How are you comparing you HTPCs, i.e. at what output resolutions? Also, are you feeding RGB or component to you PJ? I think the first point with CRT is to determine its sweet spot resolution and then work backwards from there.

It's a Barco Graphics 808 in pretty good shape (slightly worn blue tube, but the two others are good) with Roscoe color filters. I have no idea about the max resolution, but it should be a lot higher than the XGA@75 I'm feeding it now. Maybe I should go up to 1152x864@75 and test that. I'm using the special Port 3 on Barcos. It's made for computer signals, or so they tell me. The cable starts as a regular VGA plug and ends up as something that looks more like a com cable (RS-232). crtprojectors.co.uk recommends that 1152 resolution for Barcos, so it's probably worth a shot. The problem is that I probably will have to re-adjust focus for that, and I really suck, and I mean suck, at adjusting focus! :( It was adjusted for XGA by a guy from the norwegian company that imports and installs Barco projectors.

Talking about "powerful" systems, and ffdshow. I recently upgraded my HTPC and now have a 2.8G P4 800 FSB, overclocked to 3.4G. 512MB of PC4200 memory. WinXP Home tweaked for speed. I still cannot get the sort of processing out of ffdshow that others here talk of. 2 x PAL DVD res causes it to stall. Best I can do is 1440 x 540 (my pj is 960x540) and only bicubic. I use Zoomplayer with WinDVD video and AC3Filter for audio. Reclock 1.4 is in there as it gets rid of microstutters. Try Lanscos resize and even at display res of 950x540 there are stutters.

I use YV12 out, and no other processing apart from resize. All unnecessary services are off, and the PC is dedicated to DVD replay only.

I don't know how you guys do better.

now that's weird! I could run 1024x768 Lanczos on my "old" AMD XP2600 without problems. But one question really comes to the surface: Why on earth use WinXP Home? It's got nothing that Pro hasn't got, and it lacks a lot of the stuff Pro has. Like I have bragged about before I use 1920x1728 on my new HTPC without problems, with Denoise3D and dScaler sharpen before that. With that I run between 88 and 100 CPU (the spikes to 100 never shows in the picture), normally around 94 now. Can it be Reclock? I don't use that. There is at least one gremlin in your system, I can tell you! You are running without hyperthreading, I hope?

Originally posted by Mastiff
now that's weird! I could run 1024x768 Lanczos on my "old" AMD XP2600 without problems. But one question really comes to the surface: Why on earth use WinXP Home? It's got nothing that Pro hasn't got, and it lacks a lot of the stuff Pro has. Like I have bragged about before I use 1920x1728 on my new HTPC without problems, with Denoise3D and dScaler sharpen before that. With that I run between 88 and 100 CPU (the spikes to 100 never shows in the picture), normally around 94 now. Can it be Reclock? I don't use that. There is at least one gremlin in your system, I can tell you! You are running without hyperthreading, I hope?

Well Pro and Home shouldn't have any effect on system speed. If anything, Home should be faster as its less clogged down with unneeded services out of the box. "Why on earth use WinXP Home" - well, cause its substantially cheaper! The only worthwhile feature Pro really has for the HTPC crowd is remote desktop, which comes in real handy when you don't feel like powering up the projector/tv to install/configure/etc.

Originally posted by Jeraden
Well Pro and Home shouldn't have any effect on system speed. If anything, Home should be faster as its less clogged down with unneeded services out of the box. "Why on earth use WinXP Home" - well, cause its substantially cheaper! The only worthwhile feature Pro really has for the HTPC crowd is remote desktop, which comes in real handy when you don't feel like powering up the projector/tv to install/configure/etc.


Yep, thats it, XP Home is cheaper. Reclock doesn't have any impact on this, but without it there are microstutters. The setup I have is very smooth in replay. I have 10 years in IT support, so I know my stuff... this is a recent clean install after the HW upgrade, and the performance has improved approx in proportion to the speed increase (was a 2.4G overclocked to 2.8), could only do pj res resize of 960 x 540 before, and likewise had to use bicubic. Video card is a Sapphire 9000 pro.

Oh, and yes, HT is off.

gazzagazza,

I'll think about your issue, sure strikes as odd. I never ever had 9000 in my system, so I was wondering if that could be the problem. What happens when you resize to 1440x960? When your filters work well there is really no need for Reclock.

Mastiff,

I think you have some homework to do to adjust your PJ before you seriously start evaluating your setup. 1024x768 is kind of low resolution for advanced setting. You should probaly try to pop in DVE in your system just to get a feeling of how well your Barco can resolve resolution. If you can't get resolution patterns perfect down to the last detail (after tweaking the scalling, I'd stay off from sharpening and denoise while doing that), it's the PJ, not the HTPC issue. By the way, 720p DLPs come relatively cheap nowadays, at least in US.

You're probably right about his card, Vlad. The 9000 is to slow, a 9500/9600 should be the minimum. I tried an older 7000 that I have laying around, and resize was hardly working at all at XGA.

As for the projector I won't downgrade to anything digital, thankyouverymuch! ;) The BG808 is a very good projector, better than any cheaper digital, and mine is in good shape except for a slight wear on the blue tube (but according to the guy from the norwegian Barco people that's something that won't be noticeable for a long time).

But you may be right about the resolution. I have been thinking about it myself, maybe I should go up. I know that the BG808 can go a lot higher, I'm not really sure how high, but I believe it's above 1600x1200@75, so maybe I will try that PAL resolution I mentioned. I haven't gone over board with ffdshow before, so I guess I need more headroom in the desktop setting to get the real goodies out of it. I haven't got DVE, but are there any tests in AVE that will do the same for me?

No offense re: digital PJs, I just feel the urge to pass on the digital call to remaining CRT guys.

AVE (probably being Avia) should be even better. My displays never went so far ahead that I needed Avia. Try horizontal and vertical resolution patterns and if you see anything but straight lines (no chroma, just luma), you have a problem.

If you can do 1600x1200, what are you waiting for? That's almost the pixel count of 1080p resolution. Resize with Avisynth Lanczos4 to 800x600 and then with FFDShow Lanczos or Spline to 1600x1200 and you will be in heaven. At least for the resolution test don't bother with sharpen and denoise (or anything else for that matter) and you will probably feel like there is no real need for additional "stuff". You have anamorfic lens with this puppy, don't you?

Originally posted by Mastiff
You're probably right about his card, Vlad. The 9000 is to slow, a 9500/9600 should be the minimum. I tried an older 7000 that I have laying around, and resize was hardly working at all at XGA.

So the video card is part of this issue then? I thought resize happened before the video card even got involved?

I'm happy to try alternatives... so should I look for an ATI 9600 variant, or go across to Nvidia - 5700 or something. I'm not up to spending on the real expensive cards though... I live in the real world where we don't earn and spend US$!!

Anyone out there got good ffdshow resize on a 9000?

I have it working fine on a 8500 however the 8500 is technically ahead of the 9000 as it was renamed to 9200 when they moved to the 9XXX designations. I am running 1440x960 with denoise3d HQ before the resize and outputting at 1770x1000i.

Today I started checking out the differences between ffdshow using YUY2 in and YV12 out and YUY2 in and YUY2 out using DVD input.

I ran some voltage checks using a scope with WinDVD and Elecard codecs in Zoom Player to see what was going on. All ffdshow settings were set to off except resize (lanczos).

At first I checked the two codecs with ffdshow not selected in Zoom Player and found they both used black (level 16)at 0mv and white (level 235)at 714mv. Then I put ffdshow in the chain and found some differences:

When I use YUY2 in and YV12 out the output to the CRT is black at level 16 (44mv)and white at level 235 (680mv).

With YUY2 in and YUY2 out the output to the CRT is black at level 0 (0mv)and white at level 255 (714mv).

Changing the decoder or Lanczos resize made no difference in this test.

This certainly explains why I prefer YUY2 out at present since my CRT is adjusted to expect a range of 0 to 714mv to run some tests and to more easily make icm profiles. Usually I have it adjusted to expect black at 45mv and white at 655mv and have used YV12 with that.

It is relatively easy to set a CRT projector up to use either of these ranges (or for that matter the 0 to 700mv new EBU N10 standard or whatever) for the RGB inputs but if a change of coding is made in ffdshow (without readjusting your projector) then "Picture Properties" needs to be used to compensate. "Levels" is easier than Picture Properties but it needs to be checked out to be sure it is fully functional. The Levels function to move level 16 black to level 0 seems OK but the part which does the level 235 to level 255 gain change needs to be checked for proper operation.

I have attached an Avia dark needles scope trace which basically shows it to be accurate as expected.

Charlie

Originally posted by Charles Black
I just did a rough estimation by assuming that your RGB output was the the same as mine at .714v for full scale. Then I just took the ratio of your scope level of .700v and multiplied it by the ratio .700/.714*256=251. If your maximum output is different then just put your value in instead of mine.Charlie

Hey Charles, would you mind teaching me how to do the scope measurement... I have the gear, but I'm lazy... if you could tell me what test pattern I should be using (I have DVE) and what pins on the VGA to be looking at I'll have a play...

Thanks in advance...

gazzagazza,

You might try looking in ffdshow "Audio Filter Configuration" and set AC3 to disabled. I also set AVIS to disabled even though I know not why.

Unrelated issue: Attached is white needles scope graph from Avia showing all is well.

Charlie

Gazzagazza,

No problem once I know what your hookup is like. Do you just have the small DB connectors on both ends of your video cable or am I real lucky and you have BNC connectors on at least one end? Also, does your scope have delayed sweep and two external triggers connectors or something else? If your scope is a "modern" digital type it may even have an event counter which allows you to pick the line of video with precision.

Charlie

Originally posted by Charles Black
Gazzagazza,

No problem once I know what your hookup is like. Do you just have the small DB connectors on both ends of your video cable or am I real lucky and you have BNC connectors on at least one end? Also, does your scope have delayed sweep and two external triggers connectors or something else? If your scope is a "modern" digital type it may even have an event counter which allows you to pick the line of video with precision.

Charlie

Just 15pin D connectors on what I have... but I can solder etc... 'scope is a Tektronix TDS1012 2 channel digital storage...

Originally posted by gazzagazza
So the video card is part of this issue then? I thought resize happened before the video card even got involved?

I'm happy to try alternatives... so should I look for an ATI 9600 variant, or go across to Nvidia - 5700 or something. I'm not up to spending on the real expensive cards though... I live in the real world where we don't earn and spend US$!!

Anyone out there got good ffdshow resize on a 9000?

The problem is that your resized DVD image is fed to your video card. It looks like that is a bottleneck in your system, as all other parts seem to have hight bandwith. In terms of upgrade, I would go with at least 9800 Pro given the bandwith of your system. Although 9600 Pro might do the job for now it might prove to be the limitation in the future. Right now 9800 Pro is below $200 US and has significant advantage in both memory interface bandwith (256 bit vs. 128 bit) and GPU speed over 9600 Pro. ATI might implement AA and AF so you can take advantage of it at 1920x1080 resize, or they might implement sharpen via shaders. For once, you won't have to use extreme cooling to get your 9600 Pro cool so it can do VMR9 without tearing at 95% CPU utilization.

This card should probably take you through next two years. Compared to 8500, 9700 was a true revolutionary card (DX9, pixel shaders galore, 96 bit floating point precision color rendering, VMR9 support, etc.), while compared to 9800 Pro, X800 brings little but sheer speed increase in 3D operation.

Vlad, I didn't even know that anamorphic lenses excisted for CRT, I thought they were digital only. But going high doesn't scare me, so maybe I'll even try 1600x1200, I have a lot of unused memory blocks, so I can experiment. :cool:

Just stop the anamorphic lens for CRT's comments right now boyos, they do not exist.

Mastiff, I would be careful about the res you run, there's a BIG DIFFERENCE between scan and resolve. 1600x1200 will almost certainly result in a soft image, you need a good 9"pj for that.

Vlad, I'll go digital when a REALLY REALLY REALLY good projector costs less than $1500 Australian, lasts 5000+hrs with no more expense, is a true multiscan device - no conversions to 60hz, has no dodgy headache inducing colour wheel, or horrible "digital" look to it, and bulbs cost A$100.

Until then I'll stick with my obsolete dinosaur running at 1280x720 at 75hz with 1000 peak lumens, which cost roughly the price of a pair of bulbs for a digital.

I just feel the urge to resist the call to digital, something about prying the beasty from my cold dead hands...;-)

Anyway, back to the topic....


Have we sorted out what the input and output levels should be for YV12 output?? (I'm not sure how to set the input to YUV etc, I'll have to play around with that).

Keep up the good work Charles!!

gazzagazza,

There is certainly something suspect with your setup. I'm using a non-pro Radeon 9600 and a Celeron 2.0 OCed at 2.66 and can do denoise3d followed by lanczos resize at 2x with NTSC DVDs (Using Owen's settings). Even with reclock, CPU utilization is in the 80-90% per the task manager. I do recall that my Radeon LE/7200 was not as good with ffdshow, so it makes sense to suspect that your video card is the current bottleneck

So, there is hope.

Originally posted by Mark_A_W


Have we sorted out what the input and output levels should be for YV12 output?? (I'm not sure how to set the input to YUV etc, I'll have to play around with that).

Keep up the good work Charles!!

Input from all the decoders seems to be YUY2 according to ffdshow, there is no setting in ffdshow you need to alter for that... if you turn on the tray icon and look at the info screen from that when video is running you'll see it there.