Ffdshow FAQ [Text View] - Page 17

Owen

04-15-06, 08:29 PM

Ok, here you go guys.
I will restrict my discussion to FFDShow, as Media Player Classic and AVI Synth filters are separate subjects.

YV12

FFDShows filters work in YV12 for speed, so any other color format will need to be converted by FFDShow before processing.
This conversion uses CPU time, and FFDShows conversion my be less then optimal, so this conversion should be avoided
For this reason we should force the Mpeg decoder to output YV12 if possible.
Both the Dscaler 5 decoder and Nvidia decoder can do this.
FFDShow’s output color space should be forced toYV12 only to prevent conversion on output.
This saves CPU and allows the video card hardware to do the conversion the RGB.
Modern video cards are designed do the conversion with great accuracy, and with no CPU overhead.

Blur & NR

Denoise3d is a very powerful denoise tool, but it needs to be understood for best performance.
The Luma and Chroma parameters affect adjacent pixel blending or blurring.
Unless things have changed in recent builds, settings for L and C under 0.8 are ignored, and 0.0 is assumed.
Settings greater then 0.8 will cause blur, and that is not something we would want to do with DVD source, at least for Luma, so Luma should be set to 0.0 for DVD.
Chroma resolution of DVD is half that of Luma, and noise is predominantly in Chroma, For this reason we want to use a setting of 1.0, or even 2.0 for Chroma, as it reduces noise without noticeable resolution loss.
Forget about fractional settings, as they are a waste of time.
Higher settings for both Luma (1.0 - 2.0) and Chroma (3.0 - 4.0) can be useful on poor quality or very noisy source.

Time is a critical setting for noise reduction. It affects the averaging of pixels over several frames (Time).
Detail in the image does not normally change much between frames, but noise pixels do.
By averaging out the change in pixels over time we can drastically reduce noise, while leaving detail intact.
A good Time setting for DVD is 4.0, with settings of 5.0 or even 6.0 useful for very noisy source.

Your proposed settings of L=1.17, C=2.63 and T=1.17 with blur the image, and produce very little noise rejection.
Again, forget about fractional settings, just use 1.0, 2.0, 3.0.

The HQ setting is recommended, but the popular “Preview SSE2” build by Andy, has no HQ option, but has a “Fast” option in stead.
From memory, the normal mode is the same as HQ in other builds, where as “Fast” invokes hand optimized code that is VERY fast, and can be used on 1080i HD.

Sharpening

The “Sharpen” filters in FFDShow are rather harsh and ugly, and it is for that reason that I have been a stronge advocate of using the resize function to do the job.
FFDShow’s high quality scaling, and the sharpen function therein, is THE reason for most people to use FFDShow, and it has been completely overlooked in your guide.
People are free to use what ever filters they like, but I think it unwise to recommend the use of any of the filters under the “Sharpen” menu.

The resize function of FFDShow has proved to be huge success on this forum, although I have to say it took a lot of prompting to get people to even try it in the beginning.
The favored resize algorithm is Lanczos, with parameter set at 2.
Higher parameter settings induce ringing and are not recommended.
People are free use as much Luma sharpening as they see fit, although I recommend discretion, but Chroma sharpening should NOT be uses, as it only increases noise in the image, but does not improve detail or sharpness.
Remember Chroma has only half the resolution of Luma on DVD as a bandwidth conservation measure. This works ok because the loss of Chroma resolution is very difficult to notice.

There used to be a bug in the Lanczos resize that cause a green color shift if whole multiples where not used.
For this reason, it has been best practice to only use 2x or 3x (double or triple input resolution) resize settings.
Settings like 2.5x would expose the color shift bug.
This may have been fixed in current builds. Maybe someone else can confirm that.

Another useful sharpen filter is the one from the Dscaler4 program.
Some people like to add it after the resize function to bring out very fine detail.

There is also now the option of the Limited Sharpen AVI Synth script, but I won’t go into that here.

Process Whole Image

“Process Whole Image” tick box instructs FFDShow to process the black bars encoded in the image (if they exist).
This is a waste of CPU, and so should NOT be used.

Post Processing.

The “Post Processing” filter is designed to help clean up over compressed low quality video, like Divx or Xvid, and was never intended for good quality DVD video.It will generally do more harm then good to decent quality source.
It is not wise to recommend the use the post processing filter in FFDShow for DVD playback.

Deinterlacing

Unfortunately the deinterlacers in FFDShow are either not very good, or are so CPU intensive, that they are unusable, especially for 1080i.
If deinterlacing is required, it is advisable to use a decoder that supports deinterlacing in hardware. (Nvidia)
That will require the decoder to run in DXVA mode, which means that FFDShow cannot be used.
The use of FFDShow in the replay chain forces software decode mode on all decoders.
IMHO the advantages of hardware deinterlacing out way any advantages resulting from the use of FFDShow.
In fact, current video cards have improved to the point where there is no longer a pressing need for FFDShow, and many people will be quite happy without it.

Dscaler decoder is IMHO the best available for quality, but it does not support DXVA, and it only supports simple Bob or Weave deinterlacing, making it unsuited for true interlaced video source.
For progressive sources (film) it works great, and there is a special version that supports 3:2 pull down.
Most video cards require a DXVA enabled decoder, and DXVA mode to enable hardware deinterlacing, but my Nvidia 7800 can deinterlaced WITHOUT DXVA mode, using the Dscaler decoder, although not with FFDShow in the replay chain.
I cannot explain how that is possible, but it is working great for me for digital TV broadcasts.

So to round up, a basic FFDShow setup would be in order-

Codes
RAW Video, All Supported

Deinterlace (if used)

Blur & NR
Dnoise3d L=0.0, C=1.0, T=4.0, HQ (Noise reduction is optional)

AVI Synth (if used)

DScaler Filter (if used)

Resize & Aspect
Multiply by 2.000 or 3.000 (No Aspect Ratio Correction)
Lanczos (Parameter 2) with Luma sharpen only.

Output
YV12 Only.

This covers the topics people need to know, I hope it is straight forward enough.

Best of luck, and have fun.

Regards,

Owen

galileo2000

04-15-06, 08:36 PM

Owen,

Excellent, thank you very much!

I've been waiting for such thing for a long time.

I believe this should be glued to the "sticky" guide on the top of the forum.

Will try it tonight on all my HTPCs (well, on most important 3 :D ))

Jeremy Duncan

04-15-06, 09:40 PM

How's it look now Owen ?

Guides (http://www.avsforum.com/avs-vb/showthread.php?t=667061)

Owen

04-16-06, 12:10 AM

Better, but you still have Denoise3d settings listed as:

Luma 1.00, Chroma 2.00, Time 1.00, HQ Checked

This is useless, and should normally be more like:

Luma 0.00, Chroma 2.00, Time 4.00, HQ Checked.

Rather the specify a Luma sharpen setting, like 1.20, which people will take literally as being THE setting to use, you should leave it up to the user, with a comment like (User Discretion).
Luma 1.20 is very high setting for a top quality display, but would be ok on average or small one.
Rememeber that less is more with sharpening.
If you go for a dramatic effect, it can easily look overdone and unnatural.

In your MPC settings you have specified “ffdshow MPEG-4 Video Decoder” AND “ffdshow raw video filter”.

“ffdshow raw video filter” should be the only one required.

In your guide for advanced HTPC’s you have not specified the use of FFDShow resize.
Resize is THE most useful addition to any high end setup, and is a must in any guide.

It is also not a given that the AVI Synth Limited Sharpen script is better then FFDShows Resize Sharpening.
It is a matter of personal preference with the particular display in use.
In fact, may people with high end displays prefer FFDShows resize sharpening, so that should be listed as an alternative.
The same goes for hqdn3d compared to FFDShows Denoise3d.

Remember that there is no one best solution for every display, especially when it comes to sharpening.
Personal preference also plays a large role, as one mans perfect picture, can be another mans over sharpened crap.

Unfortunately people tend to take guides very literally, and don’t experiment with other filters or setting, so it is important to point out the alternatives, and encourage experimentation.
There is an expectation that there is a set of “perfect setting”, that are simply the best, but unfortunately that is definitely not the case.
There are definitely some does and don’ts, but from there on out, it’s a case of test it for your self, and see what works best for you.
What type of sharpening to use, and how much to apply is very subjective, and the same can be said about noise reduction.

Keep up the good work Jeremy, we need enthusiastic people like yourself to write up guides for the uninitiated.
There is no substitute for experience in this game, but old hands like myself, can’t be on the case all the time, and we need new blood to run with the ball. :D

Regards,

Owen

Jeremy Duncan

04-16-06, 01:09 AM

(1.) you still have Denoise3d settings listed as:

Luma 1.00, Chroma 2.00, Time 1.00, HQ Checked

This is useless

(2.) Rather the specify a Luma sharpen setting, like 1.20, which people will take literally as being THE setting to use, you should leave it up to the user, with a comment like (User Discretion).

(3.) In your MPC settings you have specified “ffdshow MPEG-4 Video Decoder” AND “ffdshow raw video filter”.

“ffdshow raw video filter” should be the only one required.

(4.) In your guide for advanced HTPC’s you have not specified the use of FFDShow resize.
Resize is THE most useful addition to any high end setup, and is a must in any guide.

(5.) it’s a case of test it for your self, and see what works best for you.

I'll answer in the order I added the numbers.

(1.) It may well be useless. But that can't be helped.

I have a meathod for troubleshooting settings.
I make a change in ffdshow, and monitor it's effect.
How I monitor it's effect is I have a calibration disk. It's called "GetGray".
I use the brightness pattern with the moving bars.

If the setting is too high.
The bars in the brightness pattern become distorted.
I isolate each setting so I know the cause of the distortion.

In the case of denoise3d.
The highest it could go was 1.17, 2.63, 1.17

Then. You made the point that only whole numbers should be used.
This showed me to use the setting: 1, 2, 1

I wish I could move time up to 5 without it distorting the brightness pattern.
But it's not gonna happen.

(2.) I also test Luma sharpen in the same way I tested denoise3d.
That's the highest it can go before it distorts the picture.

(3.) Your right.

(4.) There's only need for one sharpener. And I found resizing in avisynth to gobble up cpu cycles. And resizing in ffdshow causes the tool and task bars to show.
It's easier and more efficient to just right click the screen and choose full screen.
I only used the resharpen meathed you suggested because I understood that it's sharpener was better than the one's in the sharpen tab.

(5.) I'll point that out.

isamu

04-16-06, 02:15 AM

Owen

04-16-06, 02:31 AM

(1.) It may well be useless. But that can't be helped.

I have a meathod for troubleshooting settings.
I make a change in ffdshow, and monitor it's effect.
How I monitor it's effect is I have a calibration disk. It's called "GetGray".
I use the brightness pattern with the moving bars.

If the setting is too high.
The bars in the brightness pattern become distorted.
I isolate each setting so I know the cause of the distortion.

In the case of denoise3d.
The highest it could go was 1.17, 2.63, 1.17

Then. You made the point that only whole numbers should be used.
This showed me to use the setting: 1, 2, 1

I wish I could move time up to 5 without it distorting the brightness pattern.
But it's not gonna happen.

Unfortunately you testing methodology is flawed Jeremy.

Setting Luma to 1.0 will visibly blur the image, so 0.0 should be used on good quality source.
Setting Time to 1.0 effectively disables temporal smoothing and noise reduction.
You may as well not use Denoise3d.
L=1.0, C=2.0, T=4.0 works very well with real video, with minimal side effects, yet provides significant noise rejection.
Its not perfect, but nothing is.

(4.) There's only need for one sharpener. And I found resizing in avisynth to gobble up cpu cycles. And resizing in ffdshow causes the tool and task bars to show.
It's easier and more efficient to just right click the screen and choose full screen.
I only used the resharpen meathed you suggested because I understood that it's sharpener was better than the one's in the sharpen tab.

FFDShows resize is not just about sharpening.
Resizing or scaling up 200% or 300% in FFDShow, then down to your display resolution using you video card hardware, smoothes and cleans up the low resolution DVD image significantly, to produce a much more film like image, free of aliasing.
Sharpening is optional, and many people resize only, without sharpening, just to get the above benifit.
Resize Sharpening is very subtle, and produces minimal negative side affects at modest settings.
On a large high quality display, a Luma sharpen setting of 0.5 is plenty, but on lesser displays, a setting of 1.2 would be quite acceptable.
The various sharpen systems in FFDShow and AVI Synth work in deferent ways, and can be combined.
For example FFDShows resize sharpening can be used, followed by the DScaler 4 sharpen filter.
The Limited Sharpen script has quite a few adjustment options, and can be used before or after resizing.
Sharpening done on an up scaled image is much more subtle, but is understandably much more CPU demanding.
People who want the best results possible, use VERY fast PC’s, to give them the freedom to use advanced filter chains.

As for the resizing in ffdshow causing the tool and task bars to show, well you are doing something wrong there man.
You need to exit and restart your player after applying a resize resolution change.
Sharpen settings can be changed on the fly, but resize settings cannot.

Jeremy Duncan

04-16-06, 03:15 AM

Ok. I like the denoise3d setting.
I see a bit of distortion in the test. But nothing's perfect like you said.

I set resizing in ffdshow with Media Player Classic closed.
I tried it on my 16:10 and 4:3 ratio monitors.
What's wrong with right clicking the screen and choosing full screen ?

Jeremy Duncan

04-16-06, 03:34 AM

Please look at it again and see if I missed anything.

Guides (http://www.avsforum.com/avs-vb/showthread.php?t=667061)

Owen

04-16-06, 04:54 AM

You don’t seem to understand that resizing (up scaling) in FFDShow has nothing what so ever to do with your screen or desktop resolution, aspect ratio, or the ability to run full screen.
Your player application will always downscale the output of FFDShow to your desktop resolution, using video card hardware, and can be made to display the video full screen or in a window, in any aspect ratio desired.

We are deliberately up scaling the video as far as practical in FFDShow, well beyond the desktop resolution, and then down scaling in video hardware to our desktop resolution, simply to anti alias and clean up the low resolution DVD image.
Scaling direct to the desktop resolution does NOT achieve the same result.
FFDShows scaling quality is also better the video card hardware.

When I first proposed this up scaling-anti aliasing concept about 3 years ago, I had a lot of trouble getting people to understand or accept it. The response was usually, “why bother scaling to a higher resolution then the desktop or display” and “why scale in software when the video card will do it”.
Rather then go into a lengthy discussion on over sampling theory and scaling algorithms, my answer was to get them to simply TRY it.
There were MANY doubters, but as more and more people actually tried the concept, the word got out that it actually worked, and worked remarkable well.
Back then CPU’s where a lot less powerful then today, and resizing to double DVD resolution was a big ask. Many people did not have fast enough PC’s to be able to get full benefit out of FFDShow, but those days are over.
The benefits of FFDShow software scaling are now very well known amongst the HTPC community.
Hardware scaling in current video cards is better then it was 3 years ago, but the over sampling technique utilising FFDShows Lanczos resize algorithm, can still be very worthwhile, and cannot be ignored in any high end HTPC.

Jeremy Duncan

04-16-06, 05:38 AM

Well.

I got what your saying.
I tried 1280x720 on my xga monitor and it worked.
It resized as you said.
The tool bar and task bar were still showing though.
I showed in the guides everything I did.

Could I ask a favour of you, Owen ?
Please follow the guide. Whichever one you choose.
And identify what needs to be done to get rid of the toolbar and taskbar when you resize ?

Thanks a bunch.
As it is. Even with resizing I still have to right click the screen and choose full screen.

Owen

04-16-06, 05:51 AM

Why would you resize to 1280x720. :confused:

You normally have to switch to full screen mode to watch movies in any application, so what is the problem?
Alt Enter keys, or double left click.

It’s been a long time since I looked at MPC.
I use TT or Zoom Player.

Jeremy Duncan

04-16-06, 06:01 AM

I chose that for two reasons.
1.) It's 720 i/p
2.) And it would show if it wasn't resizing on my xga screen.

What size do you recommend I put down in the guides ?
And right clicking and choosing full screen is done even with resizing ?

Owen

04-16-06, 06:30 AM

The idea is to resize to double or triple DVD resolution.
That would be 1440x960 or 2160x1440 for NTSC DVD’s or 1440x1152 or 2160x1728 for PAL DVD’s, regardless of your desktop or screen resolution.
Only people with very fast PC’s will be able to do triple DVD res, in combination with other demanding filters.

The player will always output desktop resolution in full screen mode.
Get the idea?

Jeremy Duncan

04-16-06, 06:51 AM

Owen,

Please have another look at the guides to see if it's done.

Guides (http://www.avsforum.com/avs-vb/showthread.php?p=7484897#post7484897)

Crescent

04-16-06, 09:02 AM

How do I ensure nvidia Is outputting yv12?

Owen

04-16-06, 04:44 PM

How do I ensure nvidia Is outputting yv12?

Use regedit, and look for:

[HKEY_LOCAL_MACHINE\SOFTWARE\NVIDIA Corporation\Filters\Video]

PreferYV12 REG_DWORD 0x00000001 (1)

If this does not exist, in the Filters\Video folder, right click, and add a Dword value, name it PreferYV12, then change its value to 1.

Owen

04-16-06, 05:17 PM

Jeremy,
The Entry Level PC guide is now good, however I question why you would recommend Blend deinterlacing.

If people need deinterlacing to play non film sourced DVD’s they are better of not using FFDShow, and using a decoder that supports DXVA (Nvidia), so that hardware deinterlacing can be used.

In your guide for non entry level PC’s you are resizing in AVI Synth.
The last time I checked, FFDShows internal Lanczos resize was more efficient.
To use FFDShows internal resize, and use the Limited sharpen script after resize, you would need to move AVI Synth bellow resize in the FFDShow filter chain.

There is also the assumption the Limited sharpen faster script is superior to FFDShows Resize Lanczos sharpening.
I don’t think you will get unanimous agreement on that.

Remember that people tend to take guides VERY literally, so it should be explained that Limited sharpen is an option, not the preferred option.

High end display users will probably prefer triple resize, with a small amount of Luma sharpen, to double resize with Limited Sharpen Faster.

Jeremy Duncan

04-16-06, 05:46 PM

I can't turn off the dscaler deinterlacer.
And blend is the least cpu intensive.

In the second guide. Yes, there is no resize as in the first guide.
But limitedsharpenfaster is sharpening.
If I put in ffdshow resize I would have two sharpeners;
1.) Limitedsharpenfaster
2.) resize Luma sharpen

If I put in the ffdshow resize.
Do I use Lanczos and default luma sharpen and parameter ?
Or do I set to none instead of Lanczos ?

After you reply. I can include the info you provided about triple resize.

Crescent

04-16-06, 07:01 PM

Bouna Pasqua to all!

Thanks Owen, but nvidia decoder is reporting yuy output even after the registry change.

Owen

04-16-06, 07:06 PM

Blend Deinterlacing is VERY poor quality, and should not be used.

You can resize without sharpening.
Just leave Luma Sharpen set to 0.0.
You still need Lanczos with parameter 2.

The DScaler decoder only offers Auto, Weave or Bob.
Blend is not an option.
Weave is fine for film source, but true interlaced video will require decent adaptive deinterlacing.
Hardware pixel or motion adaptive deinterlacing is not available using the DScaler 5 decoder on most hardware

Jeremy Duncan

04-16-06, 07:59 PM

Ok. I edited the advanced guide.

There is blend available as a option.
I can't turn off the dscaler deinterlacer.
Blend is the least cpu intensive one.

Owen

04-16-06, 08:26 PM

Lookin good now Jeremy

I am not familiar with MPC, so I will download a copy, and see what is going on with that deinterlacing setting.
Neither Weave or Bob require any significant CPU, as they are not really deinterlacing in the true sense, only displaying the same data in a different ways.

It could just be that blend is the same as weave in MPC.

isamu

04-16-06, 08:53 PM

Owen...

what desktop resolution should I have my Laptop at? I always use my front projector as my "extended desktop" and drag Zoom Player over to the big screen. The front projector is at 1280x720, while the LCD screen on my laptop is at 800x600. Whenever I exit zoomplayer, it switches my laptop lcd screen to 1280x720. Should I have my front projector at a higher resolution than 1280x720 in order for me to see the picture at 1440x960?

Also, is having filters in DScaler required for it to work? I have DScaler selected but I cannot choose any DScaler filters from the drop down menu.

Owen

04-16-06, 09:09 PM

isamu

04-16-06, 09:38 PM

If your projector is 1280x720, you should set the extended desktop resolution to the projectors native resolution, so that it does not need to scale.
You can never see the picture in 1440x960, as it will be scaled down to your desktop resolution for output to the projector.

Ahh I see, so I am at least doing that right :)

Personally, I would not use extended desktop, but rather clone, or even better, single monitor mode.
Performance is usually much better in single monitor mode.

Hmmm, so wait a minute......could using extended desktop be the reason why I am getting studdery video using Lanczos? I mean for crying out loud, I have a P4 3.6Ghz w/2GB ram and ATI X800 Mobility. It's arguably the most powerful notebook you can buy!

The DScaler filters come with the DScaler 4 real time TV viewer application, NOT the DScaler 5 Mpeg decoder.

You will still need to show FFDShow where to look for the filters.

Hmmm, I'm trying to do this....I'm looking in the Plugins folder, but none of the icons in any of those FLT folders have a .dll extention. Perhaps I downloaded the wrong DScaler zip?

isamu

04-16-06, 10:02 PM

O.K. I just downloaded the TV Viewer app and was able to find the dll files. Thanks Owen :)

By the way, can you shed some light on what the difference is between having the "Process Whole Image" box checked and unchecked for these various settings? Which one would you recommend having checked?

Owen

04-17-06, 12:49 AM

By the way, can you shed some light on what the difference is between having the "Process Whole Image" box checked and unchecked for these various settings? Which one would you recommend having checked?

I quote myself from yesterday. :D

Process Whole Image

“Process Whole Image” tick box instructs FFDShow to process the black bars encoded in the image (if they exist).
This is a waste of CPU, and so should NOT be used.

Crescent

04-17-06, 06:51 AM

Could someone please explain why nvidia decoder is reporting yuy output even after the registry change?

Tukkis

04-17-06, 07:26 AM

Owen, does ffdshow auto detect black bars?

Owen

04-17-06, 07:46 AM

Could someone please explain why nvidia decoder is reporting yuy output even after the registry change?

What did it report before you changed the registry?

Owen

04-17-06, 07:46 AM

Owen, does ffdshow auto detect black bars?

Yes.

Tukkis

04-17-06, 10:32 AM

So, should it use more processing on 16x9 films than 2.35?

N3W813

04-17-06, 12:23 PM

So, should it use more processing on 16x9 films than 2.35?

Yes, ffdshow does take up more cpu usage when processing a 1.85:1 image compared to a 2.35:1. :)

Mastiff

04-17-06, 03:28 PM

No screenshots. Sorry. Don't do them since they mean nothing at all (because of the difference between what you watch them on and what the graphics card does to the image).

As for the setup it's like this:
Dscaler Sharpen 100
DeNoise 3d with 1, 3, 4 and Fast - looks good on my CRT projector
Avisynth with this script:

Import("C:\Programfiler\AviSynth 2.5\plugins\LimitedSharpen.avs")

LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=4,strength=40)

Resize 1200x1200 Lancosz 2, luma sharpen 1.20 the rest off

This looks good on my system, but again your mileage will vary (not "may", "will", since I'm pretty sure you don't have a GeForce 6600 sending to a Barco 808 with a 5 meter long Port 3 cable on a 8 feet wide Draper TabTensioned screen in a completely dark room with black ceiling and red walls and carpet. Because that's the only way to be able to use the same settings with the same result. It will always change with your display and other environment factors.

Crescent

04-17-06, 06:01 PM

What did it report before you changed the registry?

Thanks.

The same.

Tukkis

04-17-06, 06:14 PM

Yes, ffdshow does take up more cpu usage when processing a 1.85:1 image compared to a 2.35:1.

So when setting up to see how much processing your CPU can handle it's best to use 16x9?

e268

04-17-06, 08:25 PM

Owen:

Thanks you for your explanations of each of the different filters of ffdshow. Some of them I do not understand until now.

I have used ZP and Andy's preview build for the past year and half. I have also followed this thread for this period of time. But I have never seen explanation of postprocessing, so, I am glad that you have done a great job explaining these functions.

Unfortunately, I have contributed to misleading some of you in the ffdshow settings, and I apologze for it. But, I would like you to hear me out because having used Andy's built (and more or less settings in line with your recommendations), I have found that using PostProcessing and Sharpen gives me a sharper picture with less noise and much closer to HDTV from my cable company. Initially, I used PowerDVD5 with ZP, and then found that Descaler 5 gives a better picture. Recently, I have been backing up my dvd's with Xvid, hence I have moved on to Media Player Classic, a player that gives me a picture even better than Dscaler 5. Then I started to experiment with limitedsharpen, which gives me a very good sharp picture, but with a little stutter (altho my AMD64 Mobile 4000+ @ 3.0GHz is only running at about 70%). So, I experimented with PostProcessing and Sharpen (unsharpmask) and Lanczos 2 resizing, and found that the pictures is as good as if not better than I ever got. If you have some time, I would appreciated it if you can give my settings a try and see what you think. It seems that what I am getting is a little contrary to your definitions of postprocessing and sharpen; and don't get me wrong here, since I have followed this thread for a long time and know that you are one of the most respected guru here. In particular, I use MPC with ffdshow with postprocessing, unsharpmask, and lanzcos 2 resizing. The settings are detailed in my earlier post. Altho Jeromy has givin it a try (with no obvious succses as I have), but I don't understand what he is doing with his monitor.

I apologize for this long post, but I sure hope to find out if I am on a right track to some settings that others may have overlooked.

nivek7

04-23-06, 02:28 AM

Is there anyway to make ffdshow filter some videos, while not filtering others? I'm asking this because I want to filter all my movies, but I have a certain xvid-encoded video which stutters badly when filtered. Is there a way to do this without manually turning resize off/on each time?

Thanks!

Jeremy Duncan

04-26-06, 06:22 PM

LanczosResize is a resizer and LimitedSharpen is a sharpener... LimitedSharpen uses a resizer internally (LanczosResize iirc) coz it upsizes the image -> sharpens it (in a tricky way) -> downsizes it... Why the upsizing and downsizing? Because the sharpening step causes aliasing (steppy/pixelated edges) and the interpolation of the downsizing step helps to eliminate this effect! Now, as LimitedSharpen does internal resizing anyway, it offers a option to determine the output resolution (which saves you from using another resizing step afterwards) if you dont wanna keep the original resolution!

Link (http://forum.doom9.org/showthread.php?p=820093#post820093)

Since Limitedsharpen(faster) uses LanczosResize iirc by default.
Even without using dest_x= & dest_y=.
Do I need to include resizing in ffdshow or by LanczosResize / dest_x= & dest_y=,
or is using the LanczosResize in Limitedsharpen(faster) enough ?

Mr.Bitey

04-27-06, 07:51 AM

Hi Jeremy,

Summary IMO: Yes you need to resize in ffdshow unless youve got an insanly fast single CPU machine and it works without stuttering - in which case its probably CLEANER to use dest_x dest_y than re-resizing afterwards. This way your not resizing twice.

Ive done some testing on dest_x etc and found resizing by LSF dest_x etc to be SLOWER than using a resize afterwards with ffdshow/lan2 if your using MT.. (you cant use MT and LSF Dest_x etc). You cannot use MT with dest_x dest_y

Ffdshow LAN2 resize after LSF is faster than using dest_x dest_y

Dest_x and Dest_y MAY be comparatively quicker if you were resizing BEFORE running LSF (I havent tried this combo because both run too slow on my kit). I cant run LSF without MT - too slow.

Cheers,
Bitey

trueimage

05-23-06, 12:51 PM

tofu

05-23-06, 02:23 PM

Which ffdshow build(s) are best to try with an AMD Athlon 64 3200+ Venice (overclocked to 2500mhz)? I am noticing that using "conservative" settings as others suggest with the 2006-05-13 build really dont work all that well, max cpu usage and some skipping.

Also, when playing xvid files or divx, I am getting some blocking/freezing where my xbox with xbmc played them flawlessly, I know the files are good. The only reason I upgraded was for better DVD playback and the ability to play 720p xvids and 720p/1080i .ts files.

Any help would be great, there must be someone else out there with this cpu or something similar that is getting usable results.

good question. i too would like to know what the best current build for amd 64's are.

i have a hard time understanding how people with p4's @ 2ghz are running higher settings than me without lag.

i can't even run 2x resize with hqdn3d enabled on dvd's without slowing down noticably.

right now i'm running avisynth (before resize, lags too much when placed after) with the following command : LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,Smode=3,strength=60,o vershoot=7)
colormatrix()

and 2x resize.

and that is pretty much borderline pushing my system which is a

claw hammer @ 2.5ghz
gig of ram
9800xt

i'd like to atleast run blur and noise reduction or preferably hqdn3d and be able to resize at the same time without lag on dvd's...

YONEXSP

05-23-06, 02:50 PM

good question. i too would like to know what the best current build for amd 64's are.

i have a hard time understanding how people with p4's @ 2ghz are running higher settings than me without lag.

i can't even run 2x resize with hqdn3d enabled on dvd's without slowing down noticably.

right now i'm running avisynth (before resize, lags too much when placed after) with the following command : LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,Smode=3,strength=60,o vershoot=7)
colormatrix()

and 2x resize.

and that is pretty much borderline pushing my system which is a

claw hammer @ 2.5ghz
gig of ram
9800xt

i'd like to atleast run blur and noise reduction or preferably hqdn3d and be able to resize at the same time without lag on dvd's...

Ditto here, I must be doing something wrong. FFDhow is do much of a CPU hog to be useful to me to be honest. There must be a better way with Single core 3000+ and up AMD CPU's

Sinobi

05-23-06, 02:56 PM

Although AMD is a great CPU for many other things,
for HTPC the best choice is still Pentium.
Too many people have too many problems with speed with AMD and video stuff.

Henrik

trueimage

05-23-06, 03:14 PM

Perhaps, but now we have many people with AMD's. I can't beleive there isnt an AMD tweaked build or something, I mean 2500 mhz is still 2500mhz, and if my p3-733 xbox could upscale dvds and xvids to 720p then an A64 2500mhz should be able to at least do that and a little more.

Fo the amount I spent, I could have bought the toshiba HD-DVD player and some movies and kept my xbox for the other stuff. (I do want to do some recording etc so that is just an example)

The point is these are current gen cpus, there has to be something very very biased with the ffdshow coding or the way it is compiled. If I knew more on that topic, I'd jump in and help. A 2.5 GHz cpu should be able to do some basic resize and sharpening, much better than what I'm achieving.

Although AMD is a great CPU for many other things,
for HTPC the best choice is still Pentium.
Too many people have too many problems with speed with AMD and video stuff.

Henrik

tofu

05-23-06, 03:18 PM

Although AMD is a great CPU for many other things,
for HTPC the best choice is still Pentium.
Too many people have too many problems with speed with AMD and video stuff.

Henrik

hyperthreading makes THAT much of a difference?

clock for clock, my amd is significantly faster than the p4's a lot of you guys are using.

trueimage

05-23-06, 03:21 PM

hyperthreading makes THAT much of a difference?

clock for clock, my amd is significantly faster than the p4's a lot of you guys are using.

Agreed, I'm talking a 2500 mhz cpu, some guys are using 2.0 or 2.4 ghz p4s. Even the 2.8's should be 2x as fast as the AMD.

YONEXSP

05-23-06, 03:38 PM

e268

05-23-06, 09:13 PM

Which ffdshow build(s) are best to try with an AMD Athlon 64 3200+ Venice (overclocked to 2500mhz)? I am noticing that using "conservative" settings as others suggest with the 2006-05-13 build really dont work all that well, max cpu usage and some skipping.

Also, when playing xvid files or divx, I am getting some blocking/freezing where my xbox with xbmc played them flawlessly, I know the files are good. The only reason I upgraded was for better DVD playback and the ability to play 720p xvids and 720p/1080i .ts files.

Any help would be great, there must be someone else out there with this cpu or something similar that is getting usable results.

I had my AMD64 3200+ @2.5 GHz running ffdshow 20040801a_preview. This one was built by Andy and has a "Fast" box instead of "HQ" box in denoise3d. In this version, I resize up to 2.8x with lanzcos 2. (2.5x with lanzcos4). I ran it with ZP 4.01 and Powerdvd5. I have 1 GB of ram @ 430 MHz (DDR) @ 1:1 2,2,2,6 (very low latency). Also, Celestica Radeon 9600xt. CPU is at about 60%. (Come to think of it, ram speed has nothing to do with it, since at the beginning I used a 512 MB samsung low cost ram (430MB@1:1 2.5,3,3,7) So, I suggest you find Andy's built which is optimized for sse2.

I have now upgraded my cpu to AMD64 Mobile 4000+ @ 3.0GHz. I can run the latest ffdshow which really hogs the cpu (about 60% of my 3.0GHz), but I can use Media Player Classic (instead of ZP).

I am sure the AMD64 Opteron will do well.

1gkar

05-23-06, 10:09 PM

Try here for the fDDShow version talked about.

http://ffdshow.leffe.dnsalias.com/andy2222/

Don't know about how it works yet as I've just started the research. And am downloading it as I write :D

1gkar

05-23-06, 10:40 PM

e268

05-23-06, 11:00 PM

Nope. Didn't work for me. Using ZP 4.51 it caused intermittent stuttering. Once I unregistered the filter ZM was back to normal. Looks like I'll just have to continue looking.

I have an AMD 643000+ Venice CPU

Your Venice core should be able to do it (I have seen oc up to 2.7GHz). What are the settings that you use?
BTW, I did use ZP 4.51 as well. In ZP, use overlay (to lower cpu usage). Lanczos 2, no luma or chroma sharpen, and Output YV12 only. Make sure you did not choose Gradual denoise and denoise3d at the same time. What decoder do you use?

tofu

05-23-06, 11:58 PM

I had my AMD64 3200+ @2.5 GHz running ffdshow 20040801a_preview. This one was built by Andy and has a "Fast" box instead of "HQ" box in denoise3d. In this version, I resize up to 2.8x with lanzcos 2. (2.5x with lanzcos4). I ran it with ZP 4.01 and Powerdvd5. I have 1 GB of ram @ 430 MHz (DDR) @ 1:1 2,2,2,6 (very low latency). Also, Celestica Radeon 9600xt. CPU is at about 60%. (Come to think of it, ram speed has nothing to do with it, since at the beginning I used a 512 MB samsung low cost ram (430MB@1:1 2.5,3,3,7) So, I suggest you find Andy's built which is optimized for sse2.

I have now upgraded my cpu to AMD64 Mobile 4000+ @ 3.0GHz. I can run the latest ffdshow which really hogs the cpu (about 60% of my 3.0GHz), but I can use Media Player Classic (instead of ZP).

I am sure the AMD64 Opteron will do well.

you use no other filters? just resize?

e268

05-24-06, 12:06 AM

you use no other filters? just resize?

With 3200+ cpu, I only use Blur and denoise3d, resize to 2.8x, and VMR9. Looked great on my Dell 24" LCD.

skepticon

05-24-06, 01:19 PM

For what it's worth, I'm running an AMD Athlon 64 X2 4400+ w/ 2 GB of RAM.
Using ffdshow-rev2546-SSE2 (multithreaded), I'm running:

Denoise3d - Luma:0.00 - Chroma 2.00 - Time:4.00 HQ checked
Resize - 4x
Luma method: Lanczos 2 taps
Chroma method: Lanczos 2 taps
Luma sharpen: 2.00
Chroma sharpen: 0.00

Both cores are utilized, with CPU to spare (more on one than the other).
Picture quality is brilliant displayed on 46" Toshiba DLP at 720p.

Using GeForce 7900GT, Purevideo decoder, and TheaterTek

No complaints about AMD performance here.

1gkar

05-25-06, 01:39 AM

Your Venice core should be able to do it (I have seen oc up to 2.7GHz). What are the settings that you use?
BTW, I did use ZP 4.51 as well. In ZP, use overlay (to lower cpu usage). Lanczos 2, no luma or chroma sharpen, and Output YV12 only. Make sure you did not choose Gradual denoise and denoise3d at the same time. What decoder do you use?

No settings' adjustments. Just the initial install. I have also tried other versions (20051129 & 2546-qcc-4.0.3-sse-nl) which caused the loss of ZP to playback .mkv files in DD; just PLII. I had tried VMR9 renderless with YV12 colorspace as someone over at htpcnews whom I respect said to use it.

I have also been having issues getting GB-PVR to audio sync at times which is another issue but may be related.

I have tried Nvidia Trial decoders & DScaler MPG 5.006 ones.

Crescent

05-25-06, 07:13 AM

YV12 is key. Be sure you're using a decoder that can output YV12 as well as set it in Zoomplayer in the DVD section. My P4 2.53 GHZ will only play smoothly with YV12 colorspace while using resize.

e268

05-25-06, 08:58 AM

No settings' adjustments. Just the initial install. I have also tried other versions (20051129 & 2546-qcc-4.0.3-sse-nl) which caused the loss of ZP to playback .mkv files in DD; just PLII. I had tried VMR9 renderless with YV12 colorspace as someone over at htpcnews whom I respect said to use it.

I have also been having issues getting GB-PVR to audio sync at times which is another issue but may be related.

I have tried Nvidia Trial decoders & DScaler MPG 5.006 ones.

VMR9 is great, but overlay takes less horsepower, so try out overlay. Also, use YV12 as suggested in previous post.

The audio sync in GB-PVR is unrelated. I sometimes experience that while watching tv, but if I put it in timeshift mode, it is always in sync.

bedo

05-25-06, 02:25 PM

Can some of you guys try this ffdshow and let me know your results?

http://ffdshow.sourceforge.net/tikiwiki/tiki-index.php?page=Getting+ffdshow

e268

05-25-06, 03:44 PM

I am currently running this:
ffdshow-20051208-gcc4.0.2-sse-x264.nl

which is just a few days newer. I am running it at 2.5x resize at about 60% cpu (4000+ @ 3 GHz).

ServerKing

05-27-06, 04:53 AM

I have read alot of this including the facts and I am really confused. There is so much what to tweak and not too tweak can someone please help me out and find whats the best configuration for my system:

AMD64 3000+ Venice 754
1Gig Ram
Nvidia 6200 Unlocked
AOpen Sound Card Optical Out
Lots of HD Space
Hyundai Q321 32" LCD
Windows MCE 2005

I already have ffdshow installed. I was currently using VLC to play everything but have recently changed my mind. What is the best video player application for running divxed DVD's? What are the preferred ffdshow settings? What screen resolution should my LCD be set at currently 1360X768. I cannot get media player to widescreen my divx files. Media player will not run my DVD's because it says a resolution problem. I am so confused on how to get the best picture I can. Please if someone can mentor me on what settings I should have set and what I can play around with.

adude

05-28-06, 03:02 PM

Here is my setup and settings.

Opteron 146 (Single Core) o'clocked to 2.6GHz
Nvidia 7900GT
2 GB RAM
Purevideo Decoder, VMR9

FFDShow (Celtic Druid's SSE2 version) settings -
Avisynth -
ColorMatrix()
LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,smode=3,strength=100)
Resize - Lanczos4, 2.2

I get aroung 65% cpu usage. Using GotAllMedia as DVD player. I got the similar performance using Zoomplayer as well.

isamu

06-07-06, 06:35 PM

Guys. I've been watching some movies and I'm experiencing some "screen tearing" in my picture, like if it's not on vsync or somemthing. I can't describe it, but it's like there is a horizontal "tearing" in the middle of the screen during rapid camera movements. How can I eliminate this?

Mr.Bitey

06-08-06, 12:40 AM

Powerdvd7 and FFDSHOW = NO GO

Im getting graphedit errors (killing off zoom player) if I use Pdvd7 and Ffdshow together - anyone else getting this?

Pdvd7 codec plays fine by itself (in zoomplayer) but when ffdshow is added it all ends in tears.

Cheers,
Bitey

Mr.Bitey

06-08-06, 12:45 AM

iasmu,

There isnt one 'fix all' answer.. there are a lot of things you can try to help - theres lots and lots of threads about it on here..

I found that turning on all the '3d' settings on my graphics card (antialiasting and antristopic filtering - setting them to max) fixed my tearing problems when using vmr9...

Some find setting the agp bus from 8x to 4x can fix it... Some people find changing the agp apature size can make a difference (for agp video cards that is!)..

Omega drives can help some ati card (for some people)...

Unfortunatly you need to try a lot of different things :-( unless you can find someone with the same setup as yours (video card, motherboard, ram etc)

Cheers,
Bitey

adams828

06-08-06, 03:03 PM

Okay so I just got a new LCD and want to try this out. I downloaded ffdshow and MPC and followed the settings as best I could, but not sure if it made much difference.

I'm not stuck on MPC, just happened to be the 1st I downloaded.

I'm running a Pentium D 930 and Radeon X300 if that helps.

Is there a list of optimal settings I should be using?

Sinobi

06-08-06, 03:10 PM

look in the sticky's on this sub-forum

Henrik

isamu

06-12-06, 12:47 AM

The 7800 offers a slightly cleaner and more detailed image then the 6600GT, and is even smoother for 1080i playback using TheaterTek, although the 6600GT was already very good.

Full screen renderless mode is still required with a 1080i desktop.
Other desktop resolutions may be ok.

How much better is the 7800 over an ATI X800 Mobility?

isamu

06-12-06, 12:53 AM

Sorry I did not respond, somehow I passed your post up. I have blur/NR first, then sharpen (msharpen), then resize, and lastly outputted to YV12 of course.

I've never had a problem with any of the resizes or filters - maybe it is your ffdshow build.

BTW, I was first using NVPP along with ffdshow until some research in the forum warned against NVPP. I had some problems with stuttering when using both. But once I dropped NVPP, smooth sailing.

what is NVPP?

mystery

06-12-06, 06:50 AM

I think that means NVIDIA post processing.

Rydis

06-13-06, 07:01 PM

I am having problems playing dvd's in zoom player. I recently installed zoomplayer, and used the setup sticky at the top of the page and I canplay AVI files fine so far. But when I try an ISO backup or the actual dvd, I get acess violation. Then the only way to close zoomplayer is through the task manager by ending the process. Pressing the close button does nothing. Neither will minimize or anything else. I uninstalled everything, redid it all again and same problem.

http://img119.imageshack.us/img119/4450/acessvio7lj.th.jpg (http://img119.imageshack.us/my.php?image=acessvio7lj.jpg)

Edit: Reinstalled some codecs and tried again. Worked the first time. now it tells me on dvds:

Filter Connection Error
[DVD Navigato].ac3 --> [ffdshow audio decoder].in

Then goes on to say it failed to initialize DVD interface, filters needed for playback, or update DX. I am using DX 9.0c andI can play dvd's in Media player clasic.

adude

06-25-06, 03:45 PM

Rydis,

I would go to Options and see which audio and Video decoders have been selected. Make sure that these are from same provider. Also, disable the ffdshow audio decoder.

havix

06-27-06, 05:56 PM

Anyone getting incorrect aspect ratios with X264 encoded files? I seem to get the proper AR with VLC and then why I try MPC or Zoomplayer it doesn't size correctly. I've disabled resizing in ffdshow and have tried a couple different builds with no help. When I go into the info on the currently playing file it gives me the correct AR but doesn't seem to display it properly in any program I try besides VLC which bypasses ffdshow.

vapore0n

06-28-06, 07:52 AM

BoomerBrian

07-05-06, 08:14 PM

It appears I solved my problem with screen tearing. I am not sure why it solved it but when I checked "Use Exclusive Fullscreen mode with the VMR9 Renderless Video Renderer" under Options-> Filter Control in Zoom Player Pro the tearing no longer happens. The only drawback to this is now when in Full Screen mode I can't use my mouse to right click and pull up the menu. Does anyone know why this is or can explain exactly what this feature in Zoom Player is?

Here is my setup
Zoom Player Pro
Nvidia PureVideo
ffdshow

Thanks,
Brian

finlandia

07-06-06, 12:25 PM

IIRC renderless mode originally required exclusive fullscreen playback. This automatically disabled the mouse. The idea was that ZP needed total control of the display in order to properly execute renderless.

As far as I remember, anyway. That was a while ago and it's a bit fuzzy.

Ingram

07-11-06, 10:45 AM

FYI to ATI users. The latest Cat drivers 6.6 have a bug with DVD's when their upscaled x 2. You loose the VMR9 colour setting for some reason. 6.5 Work fine however.

Mr.Bitey

07-15-06, 09:50 AM

BoomerBrian,

It uses a a single flat '3d' surface to render the video on (IIRC it works like a game drawing a texture on a flat surface) so it uses the '3d' speed/features of your video card - hence less likely to tear.

The implementation of it for zoomplayer isnt a full implementation - more like a preview of a new feature that works well enough that it was included (also theatertek had it) in the 4.xx release of ZP.. Menus not drawing and staying on screen were one bug that hasnt been resolved yet. Context menus/menus generally will NOT stay visible (they are still there but get drawn over every frame - if your really good at knowing were your mouse is then you'll be able to fumble your way through as it flickers as your mouse moves).. You might have some luck pausing the video then trying to use the menu (works occasionally).. Your best off using the keyboard interface/remote control (thats what I do when I was using vmr9 renderless exclusive).

Cheers,
Bitey

BoomerBrian

07-15-06, 09:56 AM

Mr. Bitey,

Thanks for the information. I will program my remote to work with ZoomPlayer. I have ben trying Theatertek for the last week. It is a nice product. They have an OSD is full screen mode. I just don't know if I want to spend the money. I am assuming Zoom Player will have this feature soon. Of course if I program my Hauppage remote to work with ZP then I really won't care.

Mr.Bitey

07-15-06, 11:31 AM

BoomerBrian,

No probs - you can customize the keys in ZP too, if thers a conflict with something else. So long as your remote can simulate keystrokes you should be fine, otherwise check out a program called HIP - i use it, and it works really well for me.

There is a preview (preview 6 of ZP 5.00) on the inmatrix website - check it out - its got some interesting new features.

Cheers,
Bitey

Grimmy84

07-26-06, 01:03 PM

My problem is that my 16:9 plasma has 1024*768 resolution. If i resize in ffdshow on xvid i can resize 16:9 material to the 4:3 1024*768 and then send it out 1:1 pixel perfect to the plasma that will display it with correct aspect ratio since it's pixels are rectangular.

but when resizing anamorphic dvd:s the aspect ratio will not change no matter what aspect ratio the resolution i change to has! That means i will have to resize the picture again to make it more narrow using the 16:9 scaling function of media player classic. I've noticed in test patterns of resolutions (DVE) that this additonal scaling will introduce artifacts - more specicifally horizontal lines will be messed up...

How can i corret the aspect ratio within ffdshow rather than having the media player do it afterwards?

Luis Gabriel Gerena

07-29-06, 01:53 AM

Whats the latest ffdshow version compatible with avisynth and limited sharpen? The one I am using doesnt work.
Regards

arfster

07-29-06, 07:13 AM

Renderless exclusive doesn't really work very well with ZP - as others have said it's still quite beta. I find with ffdshow hard at work, tearing becomes quite a problem at times, thus renderless+exclusive becomes necessary. Media Player Classic's version doesn't really work either.

Unfortunately, only WMP and Theatertek implement it properly, and WMP has other problems with ffdshow. I ended up buying TT, which is fairly good value as it includes the Nvidia DVD decoder.

gsc

08-01-06, 02:50 AM

Is there any point in using ffdshow + TheaterTek without resizing (in other words, letting the video card upscale to the monitor's resolution)? I'm hoping to use denoise3d on compressed video, but I don't know if it's worth having ffdshow disable hardware acceleration.

Ideally I would use Lanczos in ffdshow, but I'm running a laptop that can't handle software upscaling.

Fjhdavid

08-01-06, 12:02 PM

Mars1

08-01-06, 03:44 PM

I read all the posts of this thread (6 hours), and I learned about everything (thanks to Owen).

I tested the last celt-druid version and the last Milan version (SSE-2 both case).

I use Overlay.

With a pentium M780 (2.26ghz), I am able to play with Denoise3D AFTER a Lanczos2 (1280,720) (I cropped before a PAL source to 640x360) in order to get a x2 resize).

Owen, do you finally get good values for denoise3D after resize?
and which values for L, C and T do you prefer?

(I can change C and L with these ffdshow version unlike with andy's preview one)

thanks

Fjhdavid,

Yes, you will get better results if you aply denoise 3D AFTER risize (you have updated your post and your question has disapeared).

I am not sure how and why you perform your cropping on your source, your image with a resize 2x will not look better that a non croped x2 resize downscaled by your videocard to your 1280x720 desktop/output,
however you will lost details, that even with a lanczos resize will not be able to bring out.

Owen, All,

this is a long time that I have not upgraded my ffdshow config, I was just using the folowing config:
- denoise 3D L=0 C=0 T=3 HQ-checked (that remove some little noise)
- lancsoz2 x3 resize with L0.4 C=0
on PAL and NTSC, in VRM9 TT2.4.1.4 YU2 mixing renderless fullscreen mode (not sure all exact terms, but all theatertek user will understand)

it was almost one year that my FFDSHOW config has only changed due to TT improvment with VRM9 fulscreen that brings a big step further into PQ quality improvment, and so I have just reduced my original lanczos2 L=2 to L=0.4 for having even a sharpen picture with less artefacts.

so... Owen, after so much time enjoying my dvds, I have decided to have an extensive looks on all forum to see if there was news into the PQ improvment research, and found your reply on LimitedSharpen AVISYNTH script: I have decided to have a look on it and I found that there was again some room to improve PQ!

I know that you do not consider AVISYNTH scripts to be a real part of FFDSHOW, and to be developed into this thread, but the truth is the folowing: FFDSHOW is no more the best solution to improve DVD's PQ anymore.

The folowing AVISYNTH scripts that I used inside FFDSHOW (i still need it for Theatertek), brings more quality that the ones inside FFDSHOW.
I use right now all AVISYNTH scripts:
- MT (for multi threading)
- lanczos resize(1280, 720)
- hqdn3d(1)
- limitedsharpenfaster
- colomatrix (for fixing the colorspace for VRM9 output)
so I do not use anymore any of the implemented filters of FFDSHOW, and the result is outstanding! I do not thought that there was room to improve the PQ, but the results are here:
-it removes the Halo arround sharpen object
-it sharpen the image without amplyfying the noise.

So my conclusion is:
- when FFDSHOW will integrate all the matrix algorithms that are used into those AVISYNTH scripts?

best regards,

(lastly edited because I forgot to describe the hqdn3d script into my list of script i use, i am just not sure if i run it before or after lanczos resize)

Fjhdavid

08-01-06, 08:13 PM

so are you using hqdn3d AFTER resize?

why did you say that TT is better than ZP ? for overlay? for VRM9?

What are your parameters for hqdn3d and LS?

Mars1

08-01-06, 08:27 PM

I will check it tomorrow.

for avisyth:
- hqdn(1) (that require less cpu than a 2 or 3 parameter, and is sufficient for my taste)
- LSF is x=1 y=1 s=30 without overshoot parameter
and if i remember my latest script I use, I place hqdn3d after resize.
but i got some ocasional stutter, that are for me inacceptable, and so maybe i will either try to
- overclock my p4 3.4ghz
- will perform more test to use a ffdshow resize before avisynth script (maybe I will go back on andy's ffdshow to just complete the fastest lanczos resize before AVISYNTH).

by the way in my previous ffdshow config, after the VRM9 renderless option that appears in latest version of theatertek, I got best PQ in "FFDSHOW only" by placing denoise 3D before resize, and use my CPU power on a x3 resize instead.

Mars1

08-02-06, 04:57 AM

I do not say that theatertek is better than zoom player, because I have not tested both.

Overlay vs vrm9
Prior my latest avisynth config, my test shows a better pq with vrm9 renderless.
Now my first tests shows that the difference is not visible. Extended test will confirm one or the other.

Regards,

Fjhdavid

08-02-06, 05:50 AM

My two cents after many tests on a Sanyo PLV Z3 videoprojector with bigsize image.

With the last ffdshow-SSE2 from Milan (16 july 2006 on freecodec) in the following order:

- x2 lanczos(tap 2) resize in ffdshow (L=0 and C=0)
- Blur&NR denoise 3D in ffdshow with L=1 C=2 and T=6 (I follow Owen suggestions) for AFTER denoise
- LSF (smode=3, strength=70 with no overshoot but automatic soft (soft=-1))

For my computer denoise3D and Lanczos are faster in the last ffdshow build than with avisynth

For overlay I just tweak the PC brightness (no contrast, no gamma correction). I tested them with DVE and it seems ok.

For happy dual-core owners, the MT plugin is really efficient (I saw it on another computer) and you gain 30% of speed with LSF.

all these settings look really perfect for a lot of DVD at any time.

The only things I would like to drop is: on some very moving fast and dark objet I have blocks.
Do you have any ideas about to remove them?

PS: for me the Andy ffdshow version doesen't work very well with ZP pro 4.51, sometimes it quits, sometimes I have no images...
and the problem is that there is no way to change L and C in order to denoise3D AFTER lanczos

BoomerBrian

08-02-06, 08:17 AM

BangoO

08-02-06, 08:19 AM

Hi there,

Did anyone manage to apply LimitedSharpen (or LimitedSharpenFaster) realtime on 1280*720p sources ?
I have an X2 3800+ @ 2*2500Mhz, and even using MT() did not help, I'm still a bit too short (I get around 20-22fps, so not that far from the 24fps I should get...)

Thx ! ;)

arfster

08-02-06, 08:21 AM

Anyone got a suggestion of which versions of ffdshow/avisynth/MT to use for a dualcore Pentium? I've got a standard version running at present with 3x lanc, bit of luma sharpening, blur/denoise and sometimes others, and the cores are at 40%/80%. Very occasionally with stop/starting it will settle properly at 60/60, but most of the time it's unbalanced, and after trying lots of versions I still can't find one that splits the load properly.

BangoO

08-02-06, 03:31 PM

Another question... still on 1280*720p sources, why is that when I resize to 2560*1440, I have don't have access to color control anymore ?
On TT the slide bars do nothing, and on ZP, they are all set to unsupported (brightness, contrast, hue, saturation...).
It works without a problem using a 1920*1080 resize !

PS: I tried 2 ffdshow versions, same result

Mr.Bitey

08-02-06, 09:39 PM

Fjhdavid,

denoise3d in ffdshow is not the same as hqdn3d within avisynth..

If you want to use another denoiser after LSF then put it in avisynth (after LSF)

Can you describe these blocks? are they countours of square blocks?

Boomerbrian,
Resising before LSF is the best, but you need a huge amount of CPU to pull it off :)

BangoO,
You could, but it runs like a dog on most peoples pc's - the people whom can get LSF to run after resize would be ok, but for most of us with <4Ghz machines its just not possible - its just a matter of cpu.

arfster,
There is a version of ffdshow that allows multithreading (it didnt work on my HT system tho), that would be the one to try. Alternatively you can use the MT plugin for

BangoO,
Dont know about the colour bars - try asking blight over on the inmatrix forums (for zp)

Cheers,
Bitey

Owen

08-03-06, 05:43 AM

Sorry for my lack of attendance in this thread lately people.
I am now a 1080i HD junkie, and rarely view DVD’s.

TT’s full screen renderless VMR9 rendering on my 7800 card is now so good that I find little need for FFDShow most of the time.
My display is also now very sharp and I cant tolerate much software sharpening. A small amount of Luma sharpening in Lanczos 2 is about all I can tolerate.
Limited sharpen does not do anything useful for me.

As for Denoise3d, it should be used before resize if you want the best noise rejection performance.
Be aware the “L” values greater then 0 will cause Luma detail loss.
Values of “C” greater then 0 will blur color (Chroma), but we can get away will a value of 2 as color resolution is half that of Luma in compressed video, and noise reduction performance is improved.
A good compromise seems to be L=0, C=2, T=4 to 6.

I have seen some newer builds of FFDShow that seem to have issued with Denoise3d, as I seen motion artefacts that I never remember seeing in older builds.
I don’t know if this is because of my display setup or FFDShow. I have not had the time to track the cause down definitively.

Remember, there are no hard and fast rules to FFDShow use, nor is there a perfect single setup that suites everyone.
As long as you have some understanding of how the various filters work, you can experiment to find a setup that suits your display and personal preference.

Go for it guys.

Slammy1

08-03-06, 06:04 AM

I've been using 20040801a_preview_SSE2 as I have problems with anything newer. I get artifacting on 20041012-sse2, and any of the newer versions give me GraphEdt crashes. But I've noticed with 080104 that I get some subtle pauses on older DVDs. I didn't even notice with newer DVDs until I was playing some older ones, now it stands out. My CPU usage is low, I run more relaxed adjustments than most here. I guess my system's showing it's age (P4P800-Dlx), but it'd be nice to find 1 functioning version of ffdshow.

Fjhdavid

08-03-06, 06:21 AM

Mr Bitey,

I didn't know hqdn3d was different from Denoise3d but I am not able to run hqdn3d as fast as denoise3d (due to SSE2 in ffdshow and not in avisynth maybe?)

For my block issue, I found a solution: I use Postprocessing with minimum settings:
- mplayer in custom mode with ONLY Luminance Deblock(H) and Deblock(V) and no chrominance neither Dering and I set strength down to 30% and it works.
If you know what DERING means I am intrested.

Do you have the adress where to download multithread ffdshow?

Owen

Thanks for your "before resize" denoise suggestion.

If I use postprocessing even in a very "light settings" do I have to put it before or after the denoise3d filter?

Do you have other suggestion in order to remove my block issue?

I have also a 7800 card, I downloaded TT2.4 trial and make it work with ffdshow, but I saw no differences with ZP. Why did you change?

Thanks to all

In september I will buy a intel core 2 duo 6600....and I will tell you what we can expect from these powerful processor.

BangoO

08-03-06, 06:50 AM

Mr.Bitey

08-03-06, 09:36 AM

Slammy1,

I have that mobo and use andy's sse2 preview (ffdshow-20040801a_preview_SSE2.exe).. I have issues with other builds also LOL.. theres nothing wrong with the ss2 preview - works well for me! What are these pauses you speak of? - during playback of the main title??

Fjhdavid,
hqdn3d could be seen as a better denoiser - I find it much better than the ffdshow denoise3d - seems to do a much better job of cleaning noise without adding some of the image 'lag' i often see with denoise3d... deRing is a de-ringer - used to removed/reduce excessive edge enhancement rings.. Ringing is additional lines added (or rings) next to areas of high contrast - ie. a dark building against a light sky - if there is ringing you'll see addition veritcal lines next to the transition from dark building to light background - like ripples in a pond - the worse the ringing the more 'rings' you see... note - some sharpeneres (like lanczos4) make more rings - LSF tends to not make rings... some sources have ringing in them allready (cause by edge enhancement during mastering)...

The multithreading ffdshow I believe is the following: ffdshow-20051221-gcc4.0.2-sse-x264.nl.exe
Google it. Note some antivirus programs show this as a false positive.

The multithreading part is under the 'miscalanious' then under the libravcodec - whether this has any affect on resizing (which is the biggest killer for cpu as it runs in a single thread) or not - I dont know.

Cheers,
Bitey

08-03-06, 11:48 AM

Can anyone comment on what the Core 2 Duo processors will mean for FFDSHOW? Is this sort of a no-brainer or do we even know yet?

Fjhdavid

08-03-06, 11:58 AM

Intel core 2 duo ist the new processor from Intel.

For Owen
I downloaded Elecard but I can't choose the docoder in ZP (it doesen't appear). Do I have to enter a register key?

For all

Which is the best Mpeg2 decoder for ZP and PAL film today?

thanks

PS: with dscaler 5 I have every 30 secondes a stuter, do you know how to remove this?

08-03-06, 12:03 PM

Intel core 2 duo ist the new processor from Intel.

I'm aware. This is why I asked. It may be worth waiting a few weeks for this processor on a new htpc build if someone hopes to use ffdshow. I just don't know if ffdshow can take advantage of core 2 duo like some other processors.

ste t

08-03-06, 04:21 PM

i have just done a tt system on a domestic property i am having a problem with the incoming r.c.d that is when i do a time test it does'nt trip i am using a 30ma but i know it needs to be a 100ma r.c.d in the house the r.c.d on the consumer unit will give me a time test result puzzed any ideas

Luis Gabriel Gerena

08-03-06, 04:36 PM

Have you guys tried this:
http://forum.doom9.org/showthread.php?t=104701
I found it to be a stunning sharpener since it brings out detail without ringing. You know how sometimes you are sharpening and close ups look great and then they go to a panoramic shot and the ringing and other artifacts show up...well not anymore. This thins is my favorite sharpener & cleaner ever. I was never happy with LS btw. The only catch is that it requires quite some horsepower for real time playback.
Regards

Fjhdavid

08-03-06, 06:41 PM

Luis,

I tested SeeSaw but the problem is that you need before a very strong denoiser.
Which one are you using?

Owen

I found the elecard build 2510 but I am still not able to use it with ZP.
How did you make it working?

Thanks

Luis Gabriel Gerena

08-03-06, 06:58 PM

Luis,

I tested SeeSaw but the problem is that you need before a very strong denoiser.
Which one are you using?

Thanks

This is what I use:
MT("a = last

b=a.degrainmedian(mode=2)
MT("SeeSaw(a,b, NRlimit=6, NRlimit2=7, Sstr=1.5, Slimit=5, Spower=5, Sdamplo=6, Szp=16)
colomatrix()")

You can also use Removegrain or anyother you like. I have never been a fan of denoisers but this one doesnt seem to do any visible damage but instead seems to help.

Andy Lammer

08-03-06, 07:14 PM

Thanks for the link to SeeSaw.

I just got it working with multi-threading too and will now have to experiment with the denoisers and other parameters.

- Andy

Luis Gabriel Gerena

08-03-06, 07:53 PM

Please post back on your experiences with it. Hopefully more people will try it.

Fjhdavid

08-03-06, 08:33 PM

Andy Lammer, are you the Andy of the famous ffdshow version (with fast denoise 3D)?

Andy Lammer

08-03-06, 09:41 PM

Nope, not me -> I'm not thee famous Andy who made that special ffdshow version :)

- Andy

Mr.Bitey

08-03-06, 10:01 PM

Andy,
No point hassling you for a new version then heh :-)

Others,
I didnt think seesaw would be quick enough to use as a realtime filter... I know what im doing this weekend :-)

Cheers,
Bitey

Slammy1

08-03-06, 10:04 PM

Mr.Bitey

08-03-06, 10:10 PM

Andy Lammer

08-03-06, 10:44 PM

I've been playing on my HTPC and speed doesn't seem to be an issue.
Note that you would not do a LanczosResize or it WILL clobber your HTPC.

Look at the seesaw.avs file and you will see more command line options.
There are also ssx & ssy available if you want to do supersamling like in limitedsharpen.
Didee stated that supersampling is unnecessary in his post at doom9.org
http://forum.doom9.org/showthread.php?t=104701

Why are some guys still running that old ffdshow ?
I just redid my HTPC and this afternoon downloaded the most recent of everything required, and all is A-O-K.
I'l try to ZIP up everything and place it on a FTP for download with simle install instructions.

Maybe if SeeSaw gets lots of action we could start a new thread devoted to it like was done with LimitedSharpen.

- Andy

Mr.Bitey

08-04-06, 02:18 AM

Andy,

That'd be great. If you include some sample (working) scripts too (mt and without mt) that'd be extra cool.

PS: Im running the old version of ffdshow because it just works for me! - ive tried some of the newer ones and gave up as many of them seem have have issues (like bluescreening the OS!).. Its old, but it works and im still yet to see conclusive evidence that the newer builds are quicker - the only one that might have merit is the one supposibly supporting multithreading...

Cheers,
Bitey

Ingram

08-04-06, 02:34 AM

There is a new version out today. In the resize page it has a thing called Accurate Rounding. Which is ticked. Any idea what the hell this does?

Luis Gabriel Gerena

08-04-06, 03:30 AM

Andy,
No point hassling you for a new version then heh :-)

Others,
I didnt think seesaw would be quick enough to use as a realtime filter... I know what im doing this weekend :-)

Cheers,
Bitey
Its kind of heavy compared to the LS for instance. I am currently running without any issues using ffdshow resize 2.5x lanczos2 and a little luma sharpness. Other than that, no other ffdshow except avisynth of course. Without the MT I wasnt able to run it smooth enough for real time playback and thats with an Opteron 165 at 3GHZ! So, yeah its heavy but worth the trouble indeed :D

Slammy1

08-04-06, 06:08 AM

Slammy1,

There is a physical limit on resizing with some video cards, I think its 2048x2048.. I resize to the resolution of my display (1776x1000)..

Are you using reclock?, that can fix some panning stuttering.. it could also be that MCE is using more cpu or using vmr9 rather than overlay (vmr9 is better PQ wise, but overlay is faster) ?

Cheers,
Bitey

I really do see improvement above 1920x1080 resize, though it's purely anecdotal and may have been an optimization issue when I optimized along those lines. I figured it out. I'd installed 20041012sse2 to play around and hadn't reinstalled 20040801. I was messing around with color space in ZP (RGB32) and that's where I started seeing the panning pauses. It's not near the PQ of the 1012 build, but pause-less and not too bad of quality by any stretch; plus I've been optimizing on 1012 rather than Andy's preview so hardly apples-apples.

So torn between a GPU upgrade (9700 AIW currently) or a Conroe/AM2 package. Thanks again.

arfster

08-04-06, 09:24 PM

That's really hard on the CPU. Running Seesaw with removegrain, then a *3 lanczos resize, the cores are at 50%/95%.

Looks nice though, will have to fiddle with it some.

Luis Gabriel Gerena

08-04-06, 11:28 PM

Its worth playing with it but yes its hard on the CPU. Thank God I change my previous 2.7 dual core for this 3ghz one. I love how it looks compared to anyother sharpening Ive tried.

BoomerBrian

08-05-06, 12:38 AM

I have a 3GHZ HyperThreading P4. Should I even try SeeSaw or do I not have enough juice? Seems like most people that are playing with it have Dual Cores.

Luis Gabriel Gerena

08-05-06, 12:55 AM

To be honest I see little chance of using it real time in your case but give it a try, maybe the HT could help or perhaps its better optimized for Intel...who knows. Please report on how it worked for you so people can have some feedback on different systems.

Andy Lammer

08-05-06, 09:15 AM

I spent a lot of time evaluating different SeeSaw settings, specifically 3 denoise options for it.
My HTPC is Intel 650 3.54 GHz OC'd to 4.09 GHz, 1 Gb DDR2-800 RAM, nVidia 7900GT, Theatertek v2.4.
My PJ is a JVC SX21 ( 1400x1050 ) shooting on 120: diag Vutec Silverstar screen.

My previous settings for LimitedSharpen are these:

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpenFaster.avs")
SetMTMode(2)
MT("HQDN3D(2)")
LanczosResize(1440,960)
MT("LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,Smode=3,strength=100, overshoot=7)")

I was playing with these for SeeSaw:

Import("C:\Program Files\AviSynth 2.5\plugins\seesaw.avs")
SetMTMode(2)
#a=last
#Import("C:\Program Files\AviSynth 2.5\plugins\seesaw.avs")
#SetMTMode(2)
#a=last
#b=a.HQDN3D(1)
#b=a.RemoveGrain(mode=11) #smooth
#b=a.RemoveGrain(mode=16) #medium
#b=a.RemoveGrain(mode=25) #sharp
#b=a.DegrainMedian(mode=2)
MT("SeeSaw(last, last.HQDN3D(2), NRlimit=3, Sstr=1.5, Spower=5, Sdamplo=6, Szp=16)")
#MT("SeeSaw(last, last.RemoveGrain(mode=18), NRlimit=6, Sstr=1.5, Spower=5, Sdamplo=6, Szp=16)")
#MT("SeeSaw(last, last.DegrainMedian(mode=2), NRlimit=6, Sstr=1.5, Spower=5, Sdamplo=6, Szp=16)")
LanczosResize(1440,960)

Ok, so the # are just for comments/remarks above.
I would un-# one of the 3 MT lines to make active, and view various clips over-and-over-and-over.

So, one thing I liked for all uses was doing the LanczosResize at the end.
The improvements were fairly obvious in my setup.
There is no hope-in-hell of doing a Resize before SeeSaw, as it is just too demanding.

All 3 denoise options give good results and if I had to take my pick it would be the RemoveGrain(mode=18) , though the HQDN3D(2) is a close tossup.

I understand what the goal of SeeSaw is -> not to oversharpen already sharp things and to do stronger sharpening on blurry things.

I was watching a DVD last night called "Beowulf & Grendel" and there were these viking-like guys wearing fur coats/stoles, etc sitting around a fire in the early evening, and it was windy.
It was really cool how SeeSaw could sharpen the blowing furs and display the fine detail of the hairs -> very good image depth was created.

However, going back & forth, I still prefer my LimitedSharpen over Seesaw.
What was odd in all instances of my settings & test DVD content for SeeSaw, was that the images seamed "bleached" from a colour perspective.
You'll have to forgive choice of words as I am not a video expert who can identify where the visual faults lie ..... I'm just a simple computer hack :).
Limitedsharpen had a more fluid nature to it with richer colours in comparison to SeeSaw.

I will post similarly over at Doom9 and try to get some expert feedback on my own visual oberservations.

Of course, Everyones-Milage-My-Vary depending on your hardware & software in use.

- Andy

arfster

08-05-06, 09:19 AM

Andy: don't you find using the resize within the avisynth script murders the CPU? I've had to use the ffdshow one since it's so much less demanding, but noticed a lot of folks are doing it in avisynth - am I missing something?

Phil Smith

08-05-06, 09:30 AM

I run limitedsharpen with these settings: (ss_x=1.1,ss_y=1.1,Smode=3,strength=50, overshoot=?) Not sure what my overshoot setting is. Whatever default is.

With ss_x=1.0,ss_y=1.0 settings I see little improvement. With ss_x=1.1,ss_y=1.1 settings I see a LOT, although it's all 2.8 Northwood can handle. I can't run anything else in ffdshow with those settings.

I see others mentioning settings of ss_x=1.0,ss_y=1.0 settings. Am I doing something wrong?

ORPhD

08-05-06, 09:56 AM

arfster: My experience was the same as yours. Doing the resize in avisynth was much worse from a performance standpoint than doing it in ffdshow. In fact, I can do an ffdshow Lanczos resize BEFORE LimitedSharpen with only about 70% CPU, while an avisynth Lanczos resize AFTER LS sees the CPU at 100% and stuttering. So for me, ffdshow's resize is WAY better. Ironically, this was the opposite behavior of what one of the guides (Socio's?) suggested.

Phil Smith: While ss_x/y of 1.1 is definitely better, I'm guessing most people don't use it because it has such high requirements. That's probably why you're seeing so many use 1.0 (myself included).

Luis Gabriel Gerena

08-05-06, 11:09 AM

Andy Lammer

08-05-06, 11:49 AM

AFAIK doing a Resize before LimitedSharpen AND implementing LimistedSharpen's supersampling ( having ss_xx or ss_y >1.0 ) are more-or-less accomplishing he same thing twice. So pick one or the other, but not both at the same time. ( Please correct me if my assumptions are wrong. )

FYI: In previous trials a few months ago, scaling to 1440x960 is faster than scaling to my PJ's native res of 1400x1050 -> most likely because 1440x960 is 2x DVD's native res and the math is simpler to merely x2.

I will try doing a LonczosResize in ffdshow and not avisynth and see if it is faster on my HTPC. Thanks for that tip.

Right now, pending further testing, my preferred call script is this:

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpenFaster.avs")
SetMTMode(2)
MT("HQDN3D(2)")
LanczosResize(1440,960)
MT("LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,Smode=3,strength=100, overshoot=7)")

- Andy

Luis Gabriel Gerena

08-05-06, 03:13 PM

I meant you favorite seesaw script since for me LS only creates the wavy vertical line artifact I mentioned before :-)

Phil Smith

08-05-06, 06:23 PM

AFAIK doing a Resize before LimitedSharpen AND implementing LimistedSharpen's supersampling ( having ss_xx or ss_y >1.0 ) are more-or-less accomplishing he same thing twice. So pick one or the other, but not both at the same time. ( Please correct me if my assumptions are wrong. )I understand very little about LimitedSharpen, but that hasn't stopped me from poking around and trying different things. :)

With no resize, no anything actually, other than a very basic Limited Sharpen script, I occasionally get edge jaggies on lines that are relatively vertical. The pin stripes on Zorg's vest in the 5th Element is a really good example (the scene where he's exchanging the stones for weapons). Running LimitedSharpen by itself makes those pins stripes go crazy with the jaggies. Very pronounced and easy to see. Quite annoying actually. If I run Resize before LimitedSharpen, the jaggies in the same scene all but disappear. You can only see a subtle hint of them.

I can't run Resize because it stutters badly, but you can see a definite improvement.

Andy Lammer

08-05-06, 06:33 PM

I agree, if your HTPC has enough CPU, do a resize before LimitedSharpen.
A 2x resize will create 4x the amount of data/area to further work with.

My SeeSaw script is like this:

Import("C:\Program Files\AviSynth 2.5\plugins\seesaw.avs")
SetMTMode(2)
MT("SeeSaw(last, last.HQDN3D(2), NRlimit=3, Sstr=1.5, Spower=5, Sdamplo=6, Szp=16)")
LanczosResize(1440,960)

Take a look a the SeeSaw.avs file and you will see that other parameters have defaults based on the value of parameter NRLimit.

- Andy

Phil Smith

08-05-06, 07:37 PM

I failed to mention that I'm implementing super sampling: ss_x=1.1,ss_y=1.1 . Which was bad, because that was the whole point of my post. :o

At least with my settings, running Resize and supersampling at the same does make a difference. A big difference actually. Whether that's redundant or not, I don't know. Regardless, the results are good.

gazzagazza

08-05-06, 09:08 PM

I seem to recall a discussion at one time about resize beyond a certain point and perhaps with one or other manufacturers display drivers giving rise to a colour shift due to the driver interpreting the resized image as HDTV rather than SDTV.

Can anyone comment on this?

I seem to recall a patch, or registry entry, or something to deal with this.

Thanks in advance...

arfster

08-05-06, 11:29 PM

I get that if I resize the horizontal beyond 2000 pixels, but only with Nvidia and my DVB-S software. Soon as it hits that level, ther image gets notably washed out. This is with Nvidia 81.98 drivers, 6600GT.

This doesn't happen with DVD or xvid playback - instead these crash if I go beyond something like 3000ish. Of course, that's not really much of a restriction, since that's a good way past the point it makes any visible difference :-)

BangoO

08-06-06, 04:29 AM

Andy Lammer

08-06-06, 01:28 PM

Downloads:

ffdshow latest
http://www.free-codecs.com/FFDShow_download.htm

avisynth v2.5.6a
http://sourceforge.net/project/showfiles.php?group_id=57023

avisynth plugins ( removegrain, hqdn3d, etc )
http://www.avisynth.org/warpenterprises/

Masktools v1.5.8
http://manao4.free.fr/

MT_Masktools v2.0a29
http://manao4.free.fr/

LimtedSharpenFaster.avs
- is part of the MT_Masktools v2.0a29 download above

SeeSaw.avs
http://forum.doom9.org/showthread.php?t=104701
- use 3rd download link in 1st post

ORPhD

08-06-06, 07:06 PM

AFAIK doing a Resize before LimitedSharpen AND implementing LimistedSharpen's supersampling ( having ss_xx or ss_y >1.0 ) are more-or-less accomplishing he same thing twice. So pick one or the other, but not both at the same time. ( Please correct me if my assumptions are wrong. )

My understanding is that this statement is sort of true. Although other things are also going on, for resizing purposes only, the ss_x/y parameters are similar to the resolution multiplier under ffdshow's resize. So using settings of 1.1 is equivalent to upconverting the DVD signal only to 792x528. However, all of this upsampling is subsequently lost if the dest_x/y parameters are not exclusively stated (because the default output res is always the input res). If they are stated, and upscaling is necessary to hit the target resolution (and regardless of whether ss_x/y > 1.0), Lanczos is then used to create the desired output resolution.

Since, as a general rule, I believe most (if not all) people don't currently have the CPU horsepower to supersample all the way up to their desired resolution, resize can still serve a purpose prior to Limitedsharpen, even with ss_x/y values greater than 1.0.

Hopefully this is all true and I haven't led anyone astray. Please correct me if I'm wrong.

Mr.Bitey

08-06-06, 09:28 PM

In theory dest_x/y should have resulted in faster resizing, but last time I checked, it was faster to leave dest_x/y same as input res (dont specify it) and then resize after with ffdshow.

Supersampling (the benefits of) is not lost if you do not specify dest_x/y, for example you will find that many of the edgies have been removed due to the anti-aliasing nature of scaling up, then back down.

Cheers,
Bitey

ORPhD

08-06-06, 11:56 PM

For me, it goes even a step further: it is faster for me to do the resize before LS than within LS (although I'm only going up to 1024x576).

BangoO

08-08-06, 03:30 AM

Thx Andy for the links ;)

I tried it on 1280*720p sources and it's a lot slower than LSF, and LSF was already too slow.
Maybe I did something wrong, but I doubt it... my CPU is an X2 3800+ @ 2500Mhz.
Too bad, I really wanted to test LSF and SeeSaw to compare them to an ffdshow resize&sharpen :(

Luis Gabriel Gerena

08-08-06, 03:42 AM

After trying Andys LS script and my Seesaw Im having issues finding much difference. What exactly are your cons and pros for both Andy? I havent done extensive research though so at this time they kind of seem even to me. I am currently favoring your LSF script.
Regards

Andy Lammer

08-08-06, 05:11 PM

After more playing, I still like my LimitedSharpen script over anything I seem to get out of SeeSaw.
I just see we now have a dedicated SeeSaw thread, so I will move my discussions over to there.

Some Performance FYIs for my Avisynth/LimitedSharpen:

I tried doing a resize to 1440x960 within ffdshow instead of within avisynth, and it is much slower, giving a very choppy image.

Within avisynth, I cannot do both a LanczosResize(1440,960) and try any sort of LimitedSharpen supersampling ( ss_x=1.1 ) -> just not enough horsepower in my setup.

- Andy

TforTuna

08-08-06, 05:14 PM

SeeSaw Setup and Comparison thread (http://www.avsforum.com/avs-vb/showthread.php?t=708610)

Fjhdavid

08-10-06, 06:14 AM

Dear all,

I experienced problems with Andy ffdshow 20040108_preview version:

I got the "GraphEdt.exe error" when I play PAL films.

I don't understand beacause I kept the same settings I used with other versions of ffdshow from Milan Cutka that perfectly work (denoise3D and Lanczos resize filters).
Everything is YV12.

But I would like to experiment the "fast" denoise3D from Andy.

Did you get this "GraphEdt.exe error" message and what did you do?

thanks for your help.

Francois

PS: I have this message with the Nvidia and dscaler 5 decoder.

BangoO

08-13-06, 06:02 AM

arfster

08-13-06, 08:35 AM

It's probably cos you're resizing past your video driver/card's limit. With FFDshow this often generates little bugs.

Try other driver versions maybe?

BangoO

08-13-06, 08:41 AM

Thx for the answer arfster... as Blight said, I can also use the ffdshow's color control.
I could indeed try other drivers, but before doing that, I wanted to know if anyone else noticed this with an Nvidia card (I have a 7800GTX, it used to work on my Radeon 9700Pro).

arfster

08-13-06, 11:55 AM

Not quite the same thing, but when I resize past 2000 horizontal with xvid or TV I get a brightness shift. Strangely, it happens with TV mpeg2 but not DVD - maybe to do with the lower bitrate, or the different frontend app (although that shouldn't make a difference with directx).

Ingram

08-14-06, 07:37 AM

As I said a while ago ATi Drivers past 6.5 have a bug that disables the VMR9 colour levels past a certain resolution when resizing. But no one seems to acknowledge my posts on this forum so whatever.

arayle

09-05-06, 01:40 PM

As I said a while ago ATi Drivers past 6.5 have a bug that disables the VMR9 colour levels past a certain resolution when resizing. But no one seems to acknowledge my posts on this forum so whatever.

Thanks, I will not update my ATI Drivers and keep the 6.4 for a while ;)

jmonaco

09-05-06, 01:57 PM

Does Ffdshow work in a MacOS 10.4+ environment? If yes, which version and where can I find it? If no, is there an alternate recommendation?

( I ask because I switched from a PC to Mac this past weekend for one of my computers and I love the Mac environment.)

thekochs

09-05-06, 05:34 PM

OK,

I know somewhere the was alot of discussion on what the real CPU% usage was/is. I remember to not use the FFDShow OSD but rather the Windows Task Manager. I have a 2.8Ghz P4 running HT.....my CPU% shows 45%....is this 45% or do I need to multiply x2 for 90% ? I just recall since there is effectively 2 CPUs in HT mode you have to multiply to get real usage. I ask because I have some studders occasionaly and want to know if it is most likely my CPU tapping out at 100% (45%x2+slop) which means I need to tweek my FFDShow settings or I have tons of head room with 45% and it is something else.

I'm using FFDShow 10-12-2004 SSE2 version 08.06.04

Thanks.

Slammy1

09-05-06, 06:06 PM

The %-age is based on both cores, so it is absolute, but each core should have a window to the right of the CPU usage and if you're maxing out either core you're effectively maxing out your system. You can try LSF or similar algorithm passing MT in the parameter statement to utilize both cores, though resize will still only run on 1 of the 2 cores.

Ingram, how are you seeing the Cat problems you're stating? I don't see any color issues running Cat 6.8, might just be my uneducated eye. Don't feel bad, I'm frequently ignored as well.

Speaking of ignored, I've asked this question on different threads but it seems I've stumped everyone. Why am I limited to a 2.9 resize? I have head room CPU-wise, but if I try and increase my resize to 2.95 ZPlayer crashes. I thought it was GPU, but upgrading didn't solve it. I have tried different combos of programs, always the same issues. Thanks.

thekochs

09-05-06, 06:17 PM

The %-age is based on both cores, so it is absolute, but each core should have a window to the right of the CPU usage and if you're maxing out either core you're effectively maxing out your system. You can try LSF or similar algorithm passing MT in the parameter statement to utilize both cores, though resize will still only run on 1 of the 2 cores..

Thanks, follow-up question.

I notice when I launch the movie it pegs to 100% where the studders show then over 10-15 seconds drops to 45% or so and smooth as silk. Why the spike at the beginning ? Is there a config tweek in FFDShow I need to pull back on that could effect this ?

Slammy1

09-05-06, 06:55 PM

thekochs

09-05-06, 10:04 PM

I get the same issue running LSF, but not the native denoisers. I believe it's something to do with synching, but I'm by no means an expert. You might try reclock, depending on your config,, but I'm thinking that as long as you're pegging your CPU you'll see that issue.

What is LSF ?

I've never used Reclock. I have TT2.x with XLobby front end and FFDShow. My Audigy sound card just does SP-DIF passthrough to my receiver so I've never seen an audio sync problem that I gather is why most folks use Reclock. I'm trying to keep my HTPC with as few software layers as possbile. I'll play with the Denoise setting in FFDshow and see if that makes any difference. If I don't use FFDShow and plain TT then no issues....obviously.

Thanks.

UPDATE: Well I've gone back and tried a zillon variants of my FFDShow config and no matter what movie I play for about the first 10-15 seconds it stutters with CPU use at 100%. After this time it drops to 45% and runs smooth....audio playback fine and in sync. I've tried the AVS forum post for Basic FFDShow setup which strips alot of the stuff out with just basics left. The only thing that makes a difference is turining the resize off. I have a P4 2.8Ghz and resize is 1440x960 so I would think it could handle it ? I'm assuming so because this stutter is only up front then drops off. So, any ideas on what could cause initial stutter ?

Slammy1

09-06-06, 06:03 AM

LSF=Limited Sharpen Faster. It's an alternative to the native denoiser in ffdshow. Generally looks better, but for anime type material it seems to sharpen too much for me (produces jaggies). I'm pretty close to your set up (3.0C) and resize to 2.9, it doesn't have the issue with just the native denoisers. It would seem to indicate a settings issue. Try Andy's 0801 preview build as seen here:
http://ffdshow.leffe.dnsalias.com/andy2222/

See if that solves your problem. That's the version I use, I tend to have issues with the various other builds. Also, when you uninstall the newer version of ffdshow, remember to manually delete the install directory before installing the older build.

Phil Smith

09-06-06, 09:37 AM

Generally looks better, but for anime type material it seems to sharpen too much for me (produces jaggies).If you've got enough CPU power to pull it off, try resizing before LSF. The jaggies will be greatly reduced, if not completely eliminated.

thekochs

09-06-06, 11:35 AM

Try Andy's 0801 preview build as seen here:
http://ffdshow.leffe.dnsalias.com/andy2222/

See if that solves your problem. That's the version I use, I tend to have issues with the various other builds. Also, when you uninstall the newer version of ffdshow, remember to manually delete the install directory before installing the older build.

I will do this.

I noticed on another thread your config of "I have the same system as you do (except 1.5GB KHX3500 and a 9700Pro - soon to be x850xt). I think the key is using the correct ffdshow based on other P4P800 mine and other users experience here. " Anyway, I have a Asus P4800 mobo and I ran across a old printout of a thread that Asus has a known problem in their P4x mobos for smooth DVD playback with HT enabled. On an off chance I disabled HT in my BIOS to see the effect. As background, I recently did clean/repair install of XP Pro SP2 to clean up my system. Before that I did have my system setup for non-HT and also uni-MCU. However, when I did the clean install I forgot to setup as Uni-Processor. Anyway, it was late last night when I read this old thread and I went into the BIOS and disabling HT. Now, I'm aware from other threads that there is more to it than just disabling in BIOS to go uni-MCU but oddly enough the "initial 10-15 second studder" went away and I get some micro studders after. It appears that this 10-15 second startup studder is on chapter boundries. Anyway, with HT off I looked in Task Manager and my CPU % is about 90%+. So, I'm guessing that the new studders is from this high percent and I need to go back and try changing the FFDShow items like Denoise to reduce. Strange though....I have some posts into Asus Forum to see the deal but I'm getting there is some caching or buffering going on in the Asus hardware/firmware/drivers for HT support that causes this "startup" studder. Thus, when I disable the HT support loses just enough processing to have the studders show for other reasons. Make any sense ?

Also, I asked this on AVS Forum and think I got wrong answer. Before I said that CPU showed 100% (HT enabled) then fell to 45%. They told me the 45%was true MCU use....I thought it was 45%x2=90%. Since turning off HT in BIOS the task manager just shows one MCU and since this is 90%+ I'm guessing it trully was 45%x2=90%. Thoughts ?

mpgxsvcd

09-06-06, 03:38 PM

Are there any quick tutorials that explain how to setup LSF(Limited Sharpen Faster)?

LSF=Limited Sharpen Faster. It's an alternative to the native denoiser in ffdshow. Generally looks better, but for anime type material it seems to sharpen too much for me (produces jaggies). I'm pretty close to your set up (3.0C) and resize to 2.9, it doesn't have the issue with just the native denoisers. It would seem to indicate a settings issue. Try Andy's 0801 preview build as seen here:
http://ffdshow.leffe.dnsalias.com/andy2222/

See if that solves your problem. That's the version I use, I tend to have issues with the various other builds. Also, when you uninstall the newer version of ffdshow, remember to manually delete the install directory before installing the older build.

Slammy1

09-06-06, 06:31 PM

tk, I've tried perhaps 10 ffdshow builds and always go back to Andy's preview. I know at least one other poster here with the P4P800 that came to the same conclusion about it being THE build for it. I had some issues with occasional studders with the 1012SSE2 build, close but just not as good as the preview version (I like denoisers in later versions better, but tend to have some issue whether stuuder or artifact that brings me back). I'm not sure how CPU usage is distributed with HT turned off, but with it on it tends to favor one over the other (you can set which one somehow, I forget, but I just run default). Definitely work on the preview build, I run Denoise3D (1/2/5/Fast), resize 2.9, and luma sharpen to taste (my taste is 1.2, somewhat source dependent) with vmr9 renderless. You should be able to run LSF with practically no problems. Part of the success in smooth operation is the stability of your PC. Tweaked timings, some o/c in cpu and gpu seem to go a long way. The goal of an o/c is stability first, I find with my board it's actually more stable at 218FSB than at 200 even with performace options turned off.

mpg, there's a sticky on setting up avisynth. It does a good job in getting you started, you can tweak from there. It also has links to all the relevant software, and the thread originator is very knowledgable and helpful.

PS, my processor can't handle the filters in reverse order with LSF, but I do run them as such with the native ffdshow denoiser (makes a huge difference with Ice Age, for example, but artifacts with the Futurama DVDs). Anymore I run different setups for different titles, there was a time I ran one config for all but, yes, I'm that particular (some might say, even strange). Now if I had your setup I might just max everything and be good to go no matter what I run. Ah, perhaps just a dream.

Mr.Bitey

09-06-06, 08:38 PM

Slammy1

09-06-06, 08:53 PM

Slammy1,

Ive found BIOS's newer than 1004 cause stuttering with the P4P800. Also your earlier post says "It's an alternative to the native denoiser in ffdshow." when describing LSF - it is not a denoiser, it is a sharpener. :)

Cheers,
Bitey (P4P800 Here)

That's actually a pretty impressive statement, I thought my board came with 1004 and I bought it in 5/03 (the month it came out). I've been through every BIOS since, including running a boot block from the P4P8X board to run PAT above 200FSB. Running the latest beta now (what is it, like 2 years old?). I've actually had a fair amount of success with ffdshow and this board, I'm sure it's reading your posts and your testing that's helped to avoid the pitfalls. Thanks!

I stand corrected ;).

Mr.Bitey

09-06-06, 10:06 PM

No Probs Slammy1 - that bios stutter was a _real_ pain to work out BTW: like always, I'd put a heap of 'updates' on at once (yeah I know, broke all the rules - I like living risky)...

Whats the 200FSB PAT problem? - I thaught i had pat turned on at 233FSB... maybe not - I havent been in 'there' for a while (after my last disaster - which reset my cmos and I lost my 'perfect' overclock) LOL

Bitey

thekochs

09-06-06, 10:11 PM

tk, I've tried perhaps 10 ffdshow builds and always go back to Andy's preview. I know at least one other poster here with the P4P800 that came to the same conclusion about it being THE build for it. I had some issues with occasional studders with the 1012SSE2 build, close but just not as good as the preview version.

Well I did alot of cleaning up and re-reading on AVS forum. First, I went back to what everyone said was the best performance FFDShow in Andy's 08-01-2004 Preview SSE2....but after I installed (made sure current FFDShow was uninstalled and system clean) I kept getting a execption error in the config window and it shutdown. Too bad because he had a great FAST denoise 3D algorythm in that version.

So, I went back to 10-12-2005 SSE2 one I had before. Here is waht I implemented:
Codecs: Raw Video - All Supported
Blur & NR: Denoise 3D: Luma 0, Chroma 1.1, Time 4.03, no HQ checked
Resize: 2X, Always....No Aspect Ratio Correction checked, Lancos 2 with Luma Sharpen 0.50, All others unchanged
Output: YV12

I ran this with HT disabled in BIOS but HAL still set to MP. I got 65%-75% CPU usage with spkies up to 85%.
I then changed BIOS back to HT enabled and in Task Manager set View CPU graph to combined and got 45%-65%....but honestly not sure if this is better or just reading graph is not apples-apples.

All sounds OK and image looks good but there still is a studder when the movie starts....example, when the studio name/image comes up. The CPU percent is pegged at 100% so clearly the problem. This happens in both HT & non-HT. It also happens wtih no FFDShow and just TT 2.3 but there is no studder....smooth. The falloff from 100% down for TT alone is quick.....for TT+FFDShow it takes 5-10 seconds. I noticed that once this is done I can forward to a new chapter or even go back to chapter 1 where the moive startup is and there is no studder or 100%. It like things need a few seconds to startup.

Any idea on this ? Is this a ReClock thing ?....I've never used it.

Thanks.......progess has been made.

P.S. Any thoughts on know config error for Andy's Aug-2004 version would be appreciated. I pulled a copy right from his web folder.
http://ffdshow.leffe.dnsalias.com/

Phil Smith

09-06-06, 10:56 PM

PS, my processor can't handle the filters in reverse order with LSF, but I do run them as such with the native ffdshow denoiser (makes a huge difference with Ice Age, for example, but artifacts with the Futurama DVDs). Anymore I run different setups for different titles, there was a time I ran one config for all but, yes, I'm that particular (some might say, even strange). Now if I had your setup I might just max everything and be good to go no matter what I run. Ah, perhaps just a dream.Slammy1, Actually, I do run out of CPU, and rather quickly. Resizing to 1280x720 before LSF, makes LSF process almost 3 times as many pixels, so CPU usage skyrockets. If I can learn how overclock my E6600 I should have a lot more options. Right now I'm 20% OC'd. It will do a LOT more than that--if I just knew what I was doing. :D

It dawned on me last night that SPresso was also having to process 3 times the pixels, and there was no benefit in doing so. If I run a noise filter of some sort, it would be best to do so before resizing. I'm watching DVDs that don't have noise problems, so Spresso was actually doing more harm than good. It was eliminating some fine detail.

It also seems that running better resize algorithms don't really make much difference when resizing first. Fast cubic (or something like that) seems to look as good as Lancos, and uses less CPU cycles.

I'm still feeling my way around and learning what I'm doing. So far, no matter what I try, resizing before LSF always looks a lot better than resizing after LSF.

Mr.Bitey

09-06-06, 11:02 PM

Not a reclock thing - reclock can help if you get things like stuttering on slow pans - it wouldnt make any differnce to any delays on startup.

I get a delay too when I start (but im running some avisynth scripts too).. not sure what causes it / if there is a fix.

I'd try removing ffdshow, deleting its directory, empyting the recyycle bin and running a reg cleaner, rebooting then installing a the version you want. It might also be worth checking which ffdshow.ax is registered (incase its picking up a different install).

Ive seen this problem before and its a bit of a bugger to fix (I cant remember how, I just tried a lot of different things :) Maybe even went back to an older backup/ghost image. Things can get really skrewy when you mess with lots of different versions of ffdshow.

Cheers,
Bitey

thekochs

09-06-06, 11:46 PM

I'd try removing ffdshow, deleting its directory, empyting the recyycle bin and running a reg cleaner, rebooting then installing a the version you want. It might also be worth checking which ffdshow.ax is registered (incase its picking up a different install).
Bitey

I did al this..no luck. What do you mean by "check which ffdshow.ax is registered in case its picking up a different install". What does this mean ? Where do I look ?

Also, just to be sure do you have a known good link to Andy's 08-2006 SSE2 Preview version ? I used: http://ffdshow.leffe.dnsalias.com/andy2222/

Thanks !

Slammy1

09-07-06, 06:19 AM

Hmm, I had the same problem but only with a newer version of ffdshow. I uninstalled, installed Andy's build, then uninstalled again and deleted the directory which corrected the problem (somehow). I guess I just took it for granted that it'd work for everyone. Are you sure about system stability? Since o/c'ing came up, here are some links to help in testing:
Memtest
http://www.memtest86.com/
Prime95 (checks CPU stability)
http://www.mersenne.org/freesoft.htm
CPUZ (Information gathering, will tell if PAT's enabled)
http://www.cpuid.com/cpuz.php

In o/c'ing, I run my divider high (in my case it's a 5:4) where I'm not worried about memory stability. I up my memory votage on the front end (take a look at the specs on your system memory, generally you can apply more than recommended) and target a higher CPU speed than I need. I stress the CPU for 12 hours with Prime95 (not proof of stability, but I will downclock so if it's stable for 12 hours I know it'll be stable when I down clock). Lots of people recommend testing longer in P95, but you really stress things and I'm of a mind that it harms components and that we're not looking for the maximum o/c. Watch temps closely for the 1st 30 min. to make sure you don't get thermal runaway.

Then I go to memory o/c. You need to disable the USB ports in BIOS, though perhaps that's not true on the latest version. Start with mfr recommended settings to make sure the config is stable before beginning o/c. In memtest I jump to test 5, which seems to stress things the most. If it's stable for all tests I 1 iteration I up my FSB. If not I loosen timings and try again. The principle of memory o/c is that FSB first, timings second. I'll generally back off a step or two from the max o/c indicated by this method and run memtest for a few hours (say 5 loops). I'll loosen timings on the front end to assist in the process, then optimize timings after I find speed.

Some caveats. On my board I bump VAGP as the increased bus speed puts a demand on my GPU. If you're running stock cooling I'd stick close to specs on the CPU. Lock your bus in BIOS (on the P4P800, it's PCI/AGP bus you lock) before overclocking. O/c'ing adds heat output, more important for the c and d-series processors than c2d. This increases noise with the need to run higher fan speeds so there is a balance. This is not meant to be an exhaustive overview of o/c'ing, but it should get you going but carefully monitor temps at all times. I slightly mismounted hsf could spell death for your cpu otherwise. With HT systems you used to need 2 versions of Prime95 running (the second executed through the run command, specifying affinity so you could test the second core). Not sure if the new versions the same, it's been a bit since I clocked my CPU (I only do it when it's new or as a troubleshooting step). Watch temps and watch rail voltages, With the stress to the power you could fry other components so if you see lots of fluctuations, stop and drop back to stock. When installing core software (os, a/v, etc) make absolutely certain you're stable.

Again, the goal is stability first and foremost, not max speed especially for our apps. More cpu is good, too much is bad (actually, no matter what).

thekochs

09-07-06, 07:57 AM

What does the comment abover mean "It might also be worth checking which ffdshow.ax is registered (in case its picking up a different install)."

Is there some place that FFDshow is registered in your system to check for ? I looked in registry and the FFDShow items there didn't show any revision, etc. I have Norton Systemworks and use the OBC to clean any registry problems...which I ran after FFDShow uninstall and showed no issues to clean. Is there another specific registry cleaner for FFDShow specific ?

Thanks for all the advice/help....most appreciated !!!

Note, here is one item I did do......I used the preset files from my 10-2004 SSE2 version and created new entries in Andy's. I then started modifying from there when I got the exception error and it shutdown. I assume using these is no problem ?

BoomerBrian

09-07-06, 08:57 AM

Mr.Bitey

09-07-06, 10:26 AM

thekochs

09-07-06, 04:30 PM

thekochs,

Zoomplayer has a cool filter registerer in it - it goes and finds ffdshow.ax and registeres it - the good bit is it tells you where/which one it used.

The other way is to manually reregister the ffdshow.ax file:
regsrv32 /u ffdshow.ax
regsrvr32 ffdshow.ax
do this in the ffdshow directory (where ffdshow.ax lives)

If it still doesnt work then you might be in a right pickle :-) you could try searching for ffdshow in the registry and deleting referenecs to it (you might need to re-install your view program too if it keeps references about ffdshow in the registry).

Dont use any 'tools' from any other version with andy's version - if its keeping presets then you didnt un-install the 'other' version properly - you should have any presets or settings left over if you do it right - this is probably where the error is comming from (and is why I suggested running a registry cleaner (regsupreme has a trial period))..

Cheers,
Bitey

thanks.....actually I saved the presets to a different directory to easily (I thought) restore my settings to a new version of FFDShow. I'm guessing like you that other versions don't like each others presets. I'll try to generate a preset from scratch after the clean install.....after full uninstall/delete. I'll then try the manual registering but I have a dumb question......do I go to DOS command and go to C:\Program Files\FFDShow\ and type in the below ? I ask because these don't appear to be DOS commands and I'm not sure in Windows RUN box how to get to the FFDShow directory before typing this in ? Also, is there typo in second line of regsrvr32.....is "r" before 32 exist or is it just regsrv32 ffdshow.ax ?

The other way is to manually reregister the ffdshow.ax file:
regsrv32 /u ffdshow.ax
regsrvr32 ffdshow.ax
do this in the ffdshow directory (where ffdshow.ax lives)

Thanks !

Phil Smith

09-07-06, 06:53 PM

Slammy1

09-07-06, 08:00 PM

I strongly advise against AI o/c, but if you're stable at 20% o/c you could get some major performance boosts doing things individually. Check out your Memtest stability first, you might be surprised. Be careful, I've blown 2 components by o/c'ing (one was cause I didn't lock the AGP/PCI bus... didn't have the option on my P3B-F, the other was running a 2.4c at 3.4 on stock cooler... when I upgraded my HSF the parafin was a crispy critter). If you're cautious, you'll be well rewarded, but for me I run 218 FSB even though I'm stable to at least 3.6GHz. I o/c higher in winter vs spring.

I'll help as much as I can, but c2d is not something I have any experience with. You might look up your MoBo in the appropriate subforum at Anandtech.

Mr.Bitey

09-07-06, 09:07 PM

thanks.....actually I saved the presets to a different directory to easily (I thought) restore my settings to a new version of FFDShow. I'm guessing like you that other versions don't like each others presets.

They might be ok - but I found problems going from a new version presets and installing an older version..

I'll try to generate a preset from scratch after the clean install.....after full uninstall/delete.

Check to see if it works first before putting the time in :-)

I'll then try the manual registering but I have a dumb question......do I go to DOS command and go to C:\Program Files\FFDShow\ and type in the below ? I ask because these don't appear to be DOS commands and I'm not sure in Windows RUN box how to get to the FFDShow directory before typing this in ? Also, is there typo in second line of regsrvr32.....is "r" before 32 exist or is it just regsrv32 ffdshow.ax ?

Click start, select run
type in cmd in the 'open' box
(edit to your path - you can paste it from explorer) type in: cd c:\program files\ffdshow
then you type in regsvr32 /u ffdshow.ax
then you type in regsvr32 ffdshow.ax

Sorry for the typos :-)

the dos CD command (change directory) is how you navigate to the ffdshow directory.

Cheers,
Bitey

thekochs

09-08-06, 01:34 PM

They might be ok - but I found problems going from a new version presets and installing an older version..

Check to see if it works first before putting the time in :-)

My presets work fine in the Oct-2004 SSE2 version where I had created them. What I did was save them to a MyDocs folder.....uninstalled FFDShow, deleted the Program Files directory, ran Norton Reg cleaner, rebooted. I then installed Andy's 08-2004 SSE2 Preview version, went into FFDShow config, selected "New from file" and pointed to my MyDocs saved presets. They came up and I started to edit them a little and at different times I got a "exception error" which mentioned a DLL and the FFDShow config shutdown. Was this the type error you were seeing ?

I'll try again this weekend and create a 08-2004 preset from scratch and see if this takes.

UPDATE:
Well, good news...bad news....then good. I uninstalled Oct-2004, deleted directory, clean registry, rebooted. I then installed Andy's Aug-2004 SSE2 Preview and went into config. I had NO expection error and shutdown (yeah) and entered my configs from scratch. So, for those reading....DO NOT use other versions presets with different vesions of FFDShow. I then hid all the filters except the few I wanted to enter. Below is what I put in that are active/checked

Codec: Raw Video "All Supported".......{UPDATE: See Below...must be YUY2}
Blur & NR: Denoise 3D 1.0, 2.0, 5 with FAST checked
Resize: 2X, Always, No Aspect Ratio Change Checked
Resize Settings: Lancozs 2, Luma Sharpen .50
Output: YV12

Well.....when I ran TT the screen went black after the logo. I turned FFDShow off and TT ran fine. I ran the regsvr32 commands and they went fine and still the above. I then uninstalled, cleaned reg and reinstalled TT 2.3 and no change.
After playing around with settings it took putting: Codec: Raw Video "YUY2" and not "All Supported" to get this running again. I know TT outputs YUY2 or I'd be pulling my hair out forever or bail on Andy's optimized version. I'm sure there must be a post somewhere in the 150+ FFDShow pages on this difference but it only works YUY2.

Also, GREAT news..........my CPU percent went from 50%-65% with Oct-2004 to 30%-45% on Andy's.....WOW. Also, this seemed to be enough to fix the startup Judders I saw.

Thanks for all the help !

Mark_A_W

09-08-06, 11:28 PM

Andy's version is as unstable as hell on my system (3.0 Northwood). For instance turning on Denoise3d instantly causes a Graphthingy error.

Slammy1

09-09-06, 05:11 AM

GREAT news..........my CPU percent went from 50%-65% with Oct-2004 to 30%-45% on Andy's.....WOW. Also, this seemed to be enough to fix the startup Judders I saw.

Hooray! Mr Bitey 1, computer gremlins 0. Like I always say, HTPC is not for the faint of heart.

Mark, Andy's build really seems best for the MoBo tk, mb, and myself have in common (P4P800), but the thing about ffdshow is that it's a fairly intensive program and each build has its strengths and weaknesses (or simply won't run at all...) depending on hardware. These last 3 years have seen a lot of changes in computer hardware with HT and dual core, it's hard for software developers to keep up. Open source is +/- as well; we don't have a vendor to complain to and to create patches, but otoh it is well supported by a community which has assumed ownership of the issue.

EDIT: tk, it still seems like your running a bit hot for your config. You might look into a slight o/c and tighten your memory timings. But then again, one critical rule of HTPC is if it's not broken don't fix it so if you're happy where you're at, you probably have an advantage over >50% of the HTPC community. You could eek out a few more instructions. You might also try running avisynth scripts, they definitely play a big role in my DVD viewing. I can run LSF with some lost frames at the start (more like 5 seconds rather than your stated 15) running resize a bit higher than you, but depending on the material and mastering it is the best option at times.

Mr.Bitey

09-09-06, 09:52 AM

thekochs,

Good to hear your up and running :) - its a bastard to find some of those 'wierd' things (yuy2 colourspace) in your case :) im using YV12 all through (decoder to renderer (video card does the final hardware conversion) but im not using TT (ZP here).

Each time I try another version of ffdshow it ends up being major pain to go back to andy's preview version (andy whereever you are - thanks :-).. I think im about ready to stop trying new versions and weld the side of the HTPC shut - image it and leave it alone and start enjoying some movies again! LOL

Cheers,
Bitey

Cheers,
Bitey

Phil Smith

09-09-06, 10:50 AM

Well.....when I ran TT the screen went black after the logo. I turned FFDShow off and TT ran fine. I ran the regsvr32 commands and they went fine and still the above. I then uninstalled, cleaned reg and reinstalled TT 2.3 and no change.
After playing around with settings it took putting: Codec: Raw Video "YUY2" and not "All Supported" to get this running again. I know TT outputs YUY2 or I'd be pulling my hair out forever or bail on Andy's optimized version. I'm sure there must be a post somewhere in the 150+ FFDShow pages on this difference but it only works YUY2.I didn't know TT2 output YUY2.

I had a green tinted picture problem that I fixed by selecting YUY2 output in ffdshow. This seems to indicate that my 6600GT requires a YUY2 signal. (?)

I think I have YV12 selected in Raw Video. I thought ffdshow uses YV12, and if you fed it something else, it converts it to YV12. (?)

arfster

09-09-06, 11:00 AM

It isn't picky - ffdshow can be set to only accept yv12, or yuy2 or whatever (codecs menu), and can also be told to convert/output to whatever you like(output menu). TT's purevideo codecs will default to YV12, but dscaler's default is YUY2 (can be changed in menus).

If you aren't sure what's currently in use, set the ffdshow OSD to include input and output colour space. The ffdshow OSD in general is a great bugfinder, just set it to display everything and you'll often spot problems early that way :-)

ps I've also got a 6600GT, with ffdshow set to input/ouput only YV12, and no colour problems.

Phil Smith

09-09-06, 11:05 AM

What I'm saying is I THINK ffdshow does it's work in YV12. If you feed it something else, it HAS to convert it, and that uses CPU cycles. Plus, the conversion is just another place to potentially degrade the signal.

arfster

09-09-06, 11:18 AM

Ahh, right. Yeah, certainly resize works internally in YV12, which is where the levels trick came from (outputs in YV12). Avisynth can work with others, not sure about blur/resize.

Out of curiosity, what actually doesn't use YV12 nowadays? I've got ffdshow set to accept YV12 only, and never get refused with countless codecs.

Phil Smith

09-09-06, 12:02 PM

ps I've also got a 6600GT, with ffdshow set to input/ouput only YV12, and no colour problems.I missed this.

I just upgraded my HTPC. I use to have a 6600GT AGP that worked fine with YV12. My current 6600GT PCI-E doesn't seem to like it. I have no idea why that is.

thekochs

09-10-06, 09:22 AM

I didn't know TT2 output YUY2.

I had a green tinted picture problem that I fixed by selecting YUY2 output in ffdshow. This seems to indicate that my 6600GT requires a YUY2 signal. (?)

I think I have YV12 selected in Raw Video. I thought ffdshow uses YV12, and if you fed it something else, it converts it to YV12. (?)

Yeah,

You really have to dig around the posts to find these pearls. Here is some key ones.

* TheaterTek outputs YUY2
* AVISynth requires YV12 as input. Thus means you need some null or slightly altered filter before....most have used LEVELS with just Lum changed down to 254 right before AVISYNTH.
* Most FFDShows use ALL Supported but if you have an output of YV12 then it's best to set to this for RAW Video under Codec. One example of this is if you uses the NVPP post processor in Theatertek before FFDShow. It inputs YUY2 and outputs YV12.
* Resize does work in YV12 colorspace...hence the output of YV12. You can output in RGB32, etc. but then FFDShow would do the conversion. Since most likely FFDShow is already doing one conversion to YV12 it is best to leave it up to your PJ/Display hardware to do the YV12 conversion.
* For some reason I could only get Andy's Preview version to work with YUY2 for Raw Video in Codec. I'm guessing it is not that he needs YUY2 but I'm guessing that his code needs exact match of input color space.....since I'm giving it TheaterTek then YUY2 it is.

I'm taking Owens advice and leaving FFDSgow alone. He said "less is more" and I agree. I have a 110" screen driven from my Sony PJ of 1368x768 native rez. I really like TT's quality image but I wanted to expand the native DVD 720x480 resolution to my 1368x768. I noticed if I just used TheaterTek's AR Editor or let my PJ do the scaling then the image didn't look crisp. Thus, for me I only needed FFDShow for a better scalar. My problem, like most, is to not bring my HTPC to its knees with FFDShow. My HTPC is no sloach but 2.8GhZ P4 is not bleeding edge either. The above settings really clean things up and expand. Could I improve ? Sure, you can tweek yourself into the looney house but unless you are a video freak or really understand this stuff it becomes a point of diminishing returns. IMHO.

Hope this helps others.

I sure appreciate the Forums help on all this.

thekochs

09-10-06, 09:24 AM

I missed this.

I just upgraded my HTPC. I use to have a 6600GT AGP that worked fine with YV12. My current 6600GT PCI-E doesn't seem to like it. I have no idea why that is.

Post your system specs, player setup, and also FFDShow info.

thekochs

09-10-06, 09:27 AM

Andy's version is as unstable as hell on my system (3.0 Northwood). For instance turning on Denoise3d instantly causes a Graphthingy error.

What are your FFDShow settings ?

One suggestion I have is to go through all the filters and make sure they are
turned off. I hide the ones I don't use to keep things clean.

Here is my settings in the order set. Why don't you setup a preset for this
and enter these settings.....turn off everything else and see if it works. If so,
then you can go back and turn on any other items/differences you have and see if/where the problem occurs.

Codec: Raw Video "YUY2"
Blur & NR: Denoise 3D 1.0, 2.0, 5 with FAST checked
Resize: 2X, Always, No Aspect Ratio Change Checked
Resize Settings: Lancozs 2, Luma Sharpen .50
Output: YV12

arfster

09-10-06, 09:54 AM

* TheaterTek outputs YUY2

Strange, mine doesn't. If I tell ffdshow to accept YUY2 input only, it refuses to load entirely, and its OSD is saying input from purevideo is in YV12.

thekochs

09-10-06, 10:07 AM

Strange, mine doesn't. If I tell ffdshow to accept YUY2 input only, it refuses to load entirely, and its OSD is saying input from purevideo is in YV12.

Are you just going from TT to FFDShow....nothing in between ? When I used the various other FFDShow versions ALL SUPPORTED worked, when I had NVPP before FFDShow YV12 worked (makes sense sinse NVPP is YUY2 input and YV12 output), when I went to Andy's SSE2 Preview it failed to load on ALL SUPPORTED. I'm pretty sure I tried YV12 and it also failed to load. I then stumbled across a TT Forum post on YUY2 output from TT.....note I'm using TT2.3...not TT2.4. I know the NVidia Codecs changed from those two revisions but I'm guessing the output format didn't....not sure though. Anyway, when I tried Codec YUY2 it loaded and worked.

I looked in the OSD and there is input & output colorspace I see ?

arfster

09-10-06, 12:05 PM

Yeah, too many hours of (frustrating) htpc tweaking means I always check what filters are loaded. It's source>splitter>purevideo>ffdshow>vmr9.

Colour issues are an oddity though, a lot of people seem to have problems I've never seen - maybe unique to certain hardware or drivers?

thekochs

09-10-06, 10:15 PM

If you aren't sure what's currently in use, set the ffdshow OSD to include input and output colour space. The ffdshow OSD in general is a great bugfinder, just set it to display everything and you'll often spot problems early that way

I'm sure I'm being an idiot.....too tired of looking at this stuff....but the OSD settings in Andy's 20040801a-SSE2-Preview don't seem to have input and output colorspace. What version are you running ? Am I missing something ?

Thanks.

arfster

09-10-06, 10:22 PM

I'm on a recent build, Sep06 from clsid here:

http://sourceforge.net/project/showfiles.php?group_id=173941

thekochs

09-10-06, 10:49 PM

I'm on a recent build, Sep06 from clsid here:
http://sourceforge.net/project/showfiles.php?group_id=173941

Yeah, too many hours of (frustrating) htpc tweaking means I always check what filters are loaded. It's source>splitter>purevideo>ffdshow>vmr9.

So, are you using TheaterTek ? My mind is mush tonight so dumb questions. I'm not getting your "splitter" but I don't know your hardware setup. I'm playing back from HDD ripped DVDs. Bascially Uncompressed DVD>TT2.3>FFDShow>Overlay(ATI Radeon 9600XT AGP)>DVI>DVI Sony HS20 PJ.

It looks like my FFDShow version doesn't have the input and output colorspace....I wonder if there is some other way to know for sure ?

Mr.Bitey

09-10-06, 11:24 PM

thekochs,

Are you talking about Andy's preview? - it does have options for colourspaces.. (in and out)

Cheers,
Bitey

arfster

09-10-06, 11:37 PM

So, are you using TheaterTek ?

Sorry wasn't clear: yes I use TT. The splitter I referred to is actually part of the source (MS DVD Navigator) that it uses.

If the Andy preview doesn't have input/output options in the OSD (and it is 2 yrears old, so not too surprising), you could trial and error with the input settings in the codec. If you set it to YV12 for example, it simply won't appear at all if purevideo isn't sending it that.

Output is easier, it's whatever you tell it to be. Probably best to go for YV12 only here, because the conversion to that has already been made within ffdshow, regardless of what is inputted.

thekochs

09-10-06, 11:43 PM

thekochs,

Are you talking about Andy's preview? - it does have options for colourspaces.. (in and out)

Cheers,
Bitey

In the OSD....where ? I'm missing it. I go into OSD section and don't see input or output color spaces.

FYI....I'm reading the 140+ pages FAQ and on pages 71+ there is tons of discussion on colorspaces.....wow. Anyway, I guess I'm now confused why I cannot set my input to YV12 or All Supported...and must have set to YUY2 to work with Theatertek. I found a TT Forum post that says TT outputs YUY2 which makes sense why I had to change but AVS Forum says it outputs YV12. I know 100% that when I had the Oct-2004 SSE2 build in that my input was ALL Supported and worked fine.

I think I need a drink....this stuff is driving me nuts....AND I'm happy with the fact mine works and is smooth....having to understand is killing me.

Mr.Bitey

09-11-06, 01:28 AM

thekochs

09-11-06, 01:47 AM

Under the configuration tool (in the start menu)

Under Codecs at the top (on the list on the left) - thats where you specify the input colourspace
UNder output at the bottom (of the list on the left) - is where you set the output colourspace

Input: select raw video - then the colourspace your inputting from the drop-down

Cheers,
Bitey

Sorry, I was not clear.....I meant to "display" the input and output color space in the OSD....like frame rate or resoltion, etc. is displayed. Things seem odd so I just wanted a quick way to be 100% sure that my YUY2 input and YV12 output I have set in FFDShow are really what is going on.

Anyway, after alot of digging I found some of my answers on this FAQ...page:
http://www.avsforum.com/avs-vb/showthread.php?t=246752&page=83&pp=20 where jvincent says TT outputs YUY2 in software mode and YV12 in DVXA. Now, the $100 question is I know DVXA is turned off (disabled) when FFDShow is implemented which eludes to YUY2 output. Also, I made sure my DVXA is unchecked in TT anyway...seems to be the ongoing recommendation. So, since toggling DVXA in TT toggles the hardware acceleration in NVidia decoders does this also imply/set the YUY2 or YV12 ouput ?....even if FFDShow forces this to disable ? (I'll have to try....hmmm).
Thus, to really be sure since these are the NVidia PureVideo decoders I'm wondering what the properties page says for NVidia under TT for output color space ? I did some more digging because I was wondering how to get to the NVidia Properties/Configuration since it is not loaded or accessable within TT. What I found was:

1) Click on the NVidia tray icon while TT/FFDShow Icon is running
2) Install RadLight Filter Manager v1.4 or greater, select DirectShow Filters, NVidia Video Decoder, Properties Page.
3) Or.....I guess a brute force method for the registry settings to enable are:[HKEY_LOCAL_MACHINE\SOFTWARE\NVIDIA Corporation\Filters\Video]
"EnablePropPage"=dword:00000001
"PureVideoPropPage"=dword:00000001

Even found the one for forcing YV12 output.
[HKEY_LOCAL_MACHINE\SOFTWARE\NVIDIA Corporation\Filters\Video]
"PreferYV12"=dword:00000001
{Supposively 0 means YUY2 output}

I guess I should be satisfied that mine works....but I'm not....it's driving me crazy why going from one FFDShow revision to another changed my inputs. More important is I assume if I feed FFDShow YV12 this will help my Resize function not have to do the conversion from YUY2....better CPU load.

Anyway, not at my HTPC for several days to play with this stuff.

Slammy1

09-11-06, 05:55 AM

tk, you are correct in that the OSD of Andy's preview does not allow color space display as an option.

arfster

09-11-06, 07:12 AM

More important is I assume if I feed FFDShow YV12 this will help my Resize function not have to do the conversion from YUY2....better CPU load.

Yeah, that's definitely right.

Anyway, your posts inspired me to do a few experiments. It seems the purevideo decoder and ffdshow do a bit of negotiating - if I tell ffdshow to accept YUY2 only, that's what NVidia will send it. YV12, ditto. If I say all supported, it chooses YV12.

thekochs

09-11-06, 01:27 PM

Yeah, that's definitely right.

Anyway, your posts inspired me to do a few experiments. It seems the purevideo decoder and ffdshow do a bit of negotiating - if I tell ffdshow to accept YUY2 only, that's what NVidia will send it. YV12, ditto. If I say all supported, it chooses YV12.

I wonder why on my HTPC that only YUY2 Codec Raw inputs works ? If I put All Supported or YV12 it fails to launch. Nothing special, I just have HDD DVD IFOs>TT>FFDShow Andy's 082004 SSE2 Preview>Radeon 9600XT Overlay>DVI>Sony PJ. Must be a setting somewhere I'm missing causing problems. I'm traveling until this weekend so I'm eager to try one of the below to see/force a YV12 output from PureVideo within TT. If you have the time I'd like to know if you can access this and what you see.....assumes you are using TT and didn't install PureVideo directly which I'm sure would give you the properties tab.

Also, I'm wondering if checking/un-checking DVXA in TT changes the output of PureVideo colorspace. I know this is disabled but if this is a passthrough checkbox to PureVideo perhaps this could be interesting datapoint.

Ways To See PureVideo Properties:
1) Click on the NVidia tray icon while TT/FFDShow Icon is running
2) Install RadLight Filter Manager v1.4 or greater, select DirectShow Filters, NVidia Video Decoder, Properties Page.
3) Or.....I guess a brute force method for the registry settings to enable are:[HKEY_LOCAL_MACHINE\SOFTWARE\NVIDIA Corporation\Filters\Video]
"EnablePropPage"=dword:00000001
"PureVideoPropPage"=dword:00000001

Even found the one for forcing YV12 output.
[HKEY_LOCAL_MACHINE\SOFTWARE\NVIDIA Corporation\Filters\Video]
"PreferYV12"=dword:00000001
{Supposively 0 means YUY2 output}