AVS Forum banner
Status
Not open for further replies.
1 - 20 of 89 Posts

·
Registered
Joined
·
188 Posts
Discussion Starter · #1 ·
Alright People. This is the ultimate noob question. I've spent countless time on this forum, and countless time tweaking what I have. All of that said, my PQ seems to be unchanged (and no, I'm not blind). So, I've come to the zen masters for advice. Here are the Components:


HTPC:

P4 2.4 Ghz OC'd to 3.42 Ghz

1 Gig Corsair XL Series Ram

BeyondTV for PVR App

ZoomPlayer with Sonic Codecs and WinDVD 6 for DVD Playback

FFDShow installed


Projector:

NEC LT240K: Native Res: 1024x768


Receiver:

Pioneer 1014


Comcast HD TV through Moto. 6208.


What do I need to do to in FFDShow to make PQ look best here?


As of now, all video sources (including HTPC) are run into the Pioneer 1014 with one Component to VGA output to my Projector.


I've tried each and every setting in the FFDShow settings thread (although I can't figure out how to apply them correctly), and other than the introduction of stuttering, I can't see a difference in PQ. It doesn't look bad, but I want to see the rewards of this gem of a program.


Which version of FFDShow should I use? What should I resize to? Why doesn't Lanczos seem to make a difference? Which sharpen is the right one? How exactly do I output only YV12? How much with the FFDShow with BeyondTV trick help tv (SDTV) PQ?


Please Help.


Wad
 

·
Registered
Joined
·
360 Posts
Resize as high as you can, never lower than your displays resolution.


So minimum for you would be 1024x768.

However, your PC is hella fast, so I would try 3x DVD resolution, then if that is too slow, work your way down from there.


DEFINETLY get the SSE2 version.


Your other questions: I can't help ya there. Sorry.
 

·
Registered
Joined
·
3,510 Posts
I disagree. You should resize to the native resolution of your display device.


If you resize to a higher res like 1440x960 using Lanczos, your video card will then scale that back down to the PC's desktop res of 1024x768 using Bicubic. You have just thrown away the benefit of using a higher quality resize algorithm and argueably the most important function of FFDSHOW. You also introduce more scaling artifacts by doing the operation twice.
 

·
Registered
Joined
·
2,286 Posts
There is a lot of conflicting advice being given on the resize issue... wish we could hammer this out once and for all.
Quote:
I disagree. You should resize to the native resolution of your display device.


If you resize to a higher res like 1440x960 using Lanczos, your video card will then scale that back down to the PC's desktop res of 1024x768 using Bicubic.
So, Jesse are you saying I should resize to 1386 x 788 (native on a Sony HS20) rather than the display setting of my desktop which is 1280x720? If that's the case aren't I still scaling back to 1280x720 with the bicubic? Following that logic it would seem like it would be best to resize to the display setting... of course I'm just trying to understand and that could be off.
 

·
Premium Member
Joined
·
13,880 Posts
You won't. It is ENTIRELY a personal preference thing, and will be dependent on a hundred different factors like source material, codecs, display type, etc; You need to find your own sweet spot. Blanket statements of "you should do this" when it applies to FFDshow are fairly often wrong.
 

·
Registered
Joined
·
188 Posts
Discussion Starter · #6 ·
MadPoet, I've read your posts in the massive FFDShow thread (yes, I read the whole thread). You seem fairly knowledgable.


So, given that I have provided my codecs, software, and display, could you venture a guess for 1) DVD Source Material; and 2) BeyondTV material.


I know it won't be the ideal solution, as personal preference is important, but it would give me a starting point.


Also, what about the exact steps for outputting YV12, and the other quesitons I asked.


Thanks.
 

·
Premium Member
Joined
·
13,880 Posts
Aiieee :). Ok, step 1. What video card are you using?
 

·
Registered
Joined
·
1,800 Posts
Quote:
If you resize to a higher res like 1440x960 using Lanczos, your video card will then scale that back down to the PC's desktop res of 1024x768 using Bicubic. You have just thrown away the benefit of using a higher quality resize algorithm and argueably the most important function of FFDSHOW.
Per my own experiences this is entirely wrong. I get a much improved picture going to 2016X or 1920x1920 versus 1280x720. It is not even close, especially with vmr9.


Brian
 

·
Registered
Joined
·
3,510 Posts
Take a look at these images.

resizing.zip (4.5mb)


My observations:


Resizing (resampling) up with lanczos and then back to original res with bicubic yields a virtually identical image.


Resizing up with lanczos and then back down with lanczos slightly softens the image. There really is no scenario where this operation occurs since video cards use bicubic resizing.


Resizing to a higher res, then applying unsharp mask, then resizing back to original res, results in an image that is virtually no different from one that just has unsharp mask applied without any resizing performed.



There is no "magic" in performing filter operations on an upscaled image and then resizing back to the display's target resolution. The difference is very small with the only visible difference being the amount of sharpening applied in each process. 100% unsharp mask on the 1920x1080 image is about equal to 60-65% unsharp mask applied to the unresized image. The upscaled image is not better, perhaps different slightly. It's not a coincidence that 1920 divided by 1280 = 66% and that the unsharp mask done at the higher res is ~65% of the effect of unsharp mask performed on the 1280 image.


Please feel free to show an example where resizing the kb1original.png to a higher res, then performing filtering of some sort, is superior to just doing it on the original image. Be sure that you are using the correct type of resize method for each operation and that you use lossless image compression formats to present them.
 

·
Registered
Joined
·
7,056 Posts
Jesse S has some screen shot that seem to back up common sense.


MadPoet likes what it looks like on his own display, but has yet to produce any argument more through than I like it, and the sky is blue because he painted it.


PS.

The sky is blue due to the fact that nitrogen passes light in the blue wavelength more efficiently than the anyother color, essentially filtering those colors out.
 

·
Premium Member
Joined
·
13,880 Posts
Wow... here's a thought. I like it, and I use it. As Jesse provided ZERO information about his configuration, what he used to capture the images, etc; I'm somehow supposed to fall over myself? Hey sotti, take a hint... it's not just me that likes it. THe simple fact is, I've repeatedly answered the question. It is, and always will be, a matter of taste. Nothing more, nothing less. Next time you want to carry an argument across 3 threads, why not just PM me directly?
 

·
Registered
Joined
·
3,510 Posts
What's the difference what "my configuration" is? I'm not capturing the output at my VGA port. The screenshot from Kill Bill V1 was taken from the .TS stream using hypersnap.


Resize Method:


Irfanview for resizing (resampling) with Lanczos (Photoshop 6 doesn't have it)

Photoshop 6 for resizing with bicubic (Irfanview doesn't have anything called bicubic).

Photoshop 6 for applying unsharp mask.

Images always saved in .png with no compression


There is nothing else of any relevance in "my setup". This is not a comparison of decoders or video cards. They have no bearing in the software domain where FFDSHOW operates. If you dispute my findings, use kb1original.png and prove it.
 

·
Registered
Joined
·
3,351 Posts
Quote:
Originally posted by Jesse S
What's the difference what "my configuration" is? I'm not capturing the output at my VGA port. The screenshot from Kill Bill V1 was taken from the .TS stream using hypersnap.


.... snip ....


There is nothing else of any relevance in "my setup". This is not a comparison of decoders or video cards. They have no bearing in the software domain where FFDSHOW operates. If you dispute my findings, use kb1original.png and prove it.
Your setup has everything to do with it.


For example, are you using an Nvidia or ATI card? They have subtley different scaling implementations.


Are you using Overlay or VMR? Again, the scaling implementations are different.


Are you feeding a CRT or digital (DLP/LCD)? If digital, is it 1:1 pixel mapped? If not, then you displays scaler is in the loop.


What are the particular ffdshow settings (algo, sharpness, etc) that are being used?


All of these will have an impact on what your movie looks like when it's played back. Looking at a movie is different than looking at a still image.


Noone is arguing that the image isn't changed or that the data has been altered in some way. We all know that is has.


The fact is that most people prefer the way it looks with ffdshow. It's a perception thing so comparing bits is irrelevant. And like art, beauty is in the eye of the beholder so what one person likes (very sharp, oversaturated colours) another may not (softer image, muted colours).


If you prefer one method vs. another, that's fine.
 

·
Registered
Joined
·
3,510 Posts
Quote:
For example, are you using an Nvidia or ATI card? They have subtley different scaling implementations.
ATI Radeon 9000. ATi and Nvidia both use bicubic scaling. This is irrelevant to the images I've posted.

Quote:
Are you feeding a CRT or digital (DLP/LCD)? If digital, is it 1:1 pixel mapped? If not, then you displays scaler is in the loop.
My home theater is a CRT projector. It has no native resolution but I personally use FFDSHOW resize to 1280x720 via Lanczos. That's the sweetspot of my pj and no other scaling occurs. However, I'm still comparing these images on my 19" PC CRT monitor. Irrelevant.

Quote:
All of these will have an impact on what your movie looks like when it's played back. Looking at a movie is different than looking at a still image.
DVD's are composed of ~24 still frames played per second to create motion. Using 1 still frame to examine what all this scaling/sharpening does is completely valid. Playing video will only make it harder to appraise the difference between the various processes.

Quote:
The fact is that most people prefer the way it looks with ffdshow. It's a perception thing so comparing bits is irrelevant.
So is it fact or perception? I think much of this is lack of understanding. Higher upsampled resolutions sound better but nobody is doing any indepth testing to confirm that it really works. Look at the images and tell me that upscaling+sharpening works better than just upscaling.


How many people here are even aware that if you resize to a res above the desktop res, that your video card just scales it back bicubically before outputting to VGA/DVI/component adaptor?


How many people here are aware that chroma sharpening does virtually nothing? Picture detail is mainly contained in the luma channel. Drag that chroma sharpen slider back and forth and you'll see it does virtually nothing. And yet, I see people saying they use "1.0 chroma sharpening" as if it does anything.


Take kb1original.png from the zip I posted and try out some stuff on it. Try resizing, sharpening, various combinations and types. See what actually happens.


I hope everyone commenting in this thread has actually looked at the images to see the results. If you didn't, you're just spouting rhetoric. If you think I'm "cheating", feel free to do the same operations yourself and post the images.
 

·
Registered
Joined
·
4,351 Posts
Thanks Jesse. A very good starting point.


At the risk of offending, one theory could be that FFDShow scaling just plain has a bug in it that the downscale step 'covers-up'. Can you try using ffdshow (& hypersnap) to do the lanczos upscale?


madpoet, you're certainly welcome to do otherwise, but it would be quite interesting if you played along and were able to submit screen captures from your setup.


Andy K.
 

·
Registered
Joined
·
6,653 Posts
Since I have limited power (see ffdshow settings thread) I go for a minimalist approach. I also now subscribe to the theory of the less amount of scaling steps, the better...UNLESS...see below.


I resize to my display rez of 1280x720p, lanzcos 4, then unsharp mask=13...that's it.

This gives me excellent PQ with TT 2.0 & CPU load is ~85%.


HOWEVER, I do see the reasoning for those with more powerful systems...if they resize to very high resolutions first, the other filters get applied to that many more pixels before being downsized by the video card on output resolution.
 

·
Registered
Joined
·
3,351 Posts
Quote:
Originally posted by Jesse S


Take kb1original.png from the zip I posted and try out some stuff on it. Try resizing, sharpening, various combinations and types. See what actually happens.


I hope everyone commenting in this thread has actually looked at the images to see the results. If you didn't, you're just spouting rhetoric. If you think I'm "cheating", feel free to do the same operations yourself and post the images.
I'm not in a quoting mood, so I won't bother with quoting all.


Yes, I will agree, that the upscaled and subsequently downscaled images are nearly identical. Did you notice the word "nearly". That's the whole point.


I don't see how the points you referred to are irrelevant. I watch movies on my HDTV that is connected to my video card. Pretty much everyone agrees that ATI and Nvidia cards are different. You'll get arguements as to which is better, but they are definitely different.


I like the ways movies look after my particular ffdshow processing better. There can be no arguement to that.


If you have a different preference, then that's fine too. Trying to say that one image is "better" than another by any quantitative means is just a waste of time.
 

·
Registered
Joined
·
3,510 Posts
Originally posted by Jesse S

ATI Radeon 9000. ATi and Nvidia both use bicubic scaling. This is irrelevant to the images I've posted.


VLAD: Your video card has most to do with the image quality. With Radeon 9000, you should not be even posting your results. This card is inferior to integrated chipsets on new mobos. I rarely post so blunt, but in your case it is quite warranted.



The captures and processed images are not from that machine as I clearly stated. It really makes no difference because I am capturing from the frame buffer not the VGA output. You obviously didn't read or don't understand. FYI this machine has a Radeon 9800 Pro. But I suppose that rubbish too and only someone with a Quadro or 5700/6800 is qualified to say anything.

Quote:
VLAD: More than you think. It's all over FFDShow thread. However, there is more magic to video card scaling than bicubic resize on advanced systems. You might want to do some more research.
Ok, what is a better resize algo than Lanczos?

Quote:
You just need to upgrade your HTPC system to get acces to advanced features that one can use on HTPC nowadays.
Such as? You aren't backing up anything you say with information.
 

·
Registered
Joined
·
1,358 Posts
Even before your post I deleted mine and don't intend to debate. I simply don't agree with your conclusions based on my own experience with my system and many others that I tried. There is plenty of material here on the forum that could help you if you are interested in improving image quality of your HTPC. Also remember that all depends on the particular system. So what doesn't work for you perhaps works for somebody else.
 
1 - 20 of 89 Posts
Status
Not open for further replies.
Top