Program to Compare Video Encode Quality

[rack04]

Originally Posted by LoRd_MuldeR

Try this in Avisynth:

clip1 = DirectShowSource("C:\MyEncode1.foo", audio=false)
clip2 = DirectShowSource("C:\MyEncode2.foo", audio=false)

StackVertical(clip1,clip2)

Just to add on this script. I always add a subtitle so that I know what I'm comparing.

For example:

source1 = DirectShowSource("C:\Personal\Videos\Van Helsing DXVA-HD-FAST.mp4",fps=23.9759856527702,audio=false)
source2 = DirectShowSource("C:\Personal\Videos\Van Helsing SA-HD-DVD.mp4",fps=23.9759856527702,audio=false)

StackVertical(source1.subtitle("DXVA-HD-FAST"),source2.subtitle("SA-HD-DVD"))

[Avenger007]

Is it just me or does anyone else notice MPC uses TV (16-235) color levels for playing avs files instead of PC (0-255) colors? This results in the quality of the images being lower than they actually are.
However, MPC uses PC (0-255) for all media files, so why not for avs files?

[Ranguvar]

If MPC uses a renderer other than Haali's or the Overlay Mixer, if the decoder is outputting non-RGB video, the color levels will be limited.

Either use one of those renderers, or force ffdshow to output RGB32/RGB24 video only, with high quality YV12 -> RGB conversion enabled, and the RGB Conversion set to Full Range.

[Avenger007]

Originally Posted by Ranguvar

If MPC uses a renderer other than Haali's or the Overlay Mixer, if the decoder is outputting non-RGB video, the color levels will be limited.

Either use one of those renderers, or force ffdshow to output RGB32/RGB24 video only, with high quality YV12 -> RGB conversion enabled, and the RGB Conversion set to Full Range.

If that was the case then media files (avi, wmv, mp4,..) would use TV levels to.
BTW, both Haali's and the Overlay Mixer also cause media files to use TV levels, so I'm using VMR9 (renderless). Also, all media files and avs files use PC levels in Windows Media Player.

Can you or anyone else confirm my observations?

[Sharktooth]

it's all ok here.

[Sharktooth]

Originally Posted by Poopoo

Thanks for the solution Mulder,

too bad some guys" Ego's increase with their postcount.

better speak for yourself.
as this is a public forum, everyone can reply with a valid answer.
what the original poster asked in the first post does not exist but since he was looking for some other thing instead (his request was just badly worded) he got some valid replies from others.
now, if you're finished with insults/flames, since now we all know what the original poster wants, we can help him.

[Guest]

Gents, please stay on topic. Any further gratuitous ad hominems will lead to strikes (attention: Poopoo!). Thank you.

[Avenger007]

Originally Posted by Sharktooth

it's all ok here.

Thanks for the response.
I created a thread with the problem here: http://forum.doom9.org/showthread.php?t=139623

WRT the topic, I use MPC to compare frames by removing the status and menu bars using ctrl+1, ctrl+2, ctrl+5 and ctrl+0 so the image fills the entire window.

[vmrsss]

Hi everybody. I dont think the original question in ill-posed, at the outset only eyes can compare quality, but I'd say that one of the ideas is to identify better and better metrics that approximate that comparison; otherwise most of the developments described in this forum would be pointless...

[Zarxrax]

I like to use AvsP for comparing clips. It's tabbed interface lets you easily load up a lot of avisynth scripts and then change between them using the number keys. Very useful if you have more than 2 clips to compare.

[Sharktooth]

Originally Posted by vmrsss

Hi everybody. I dont think the original question in ill-posed, at the outset only eyes can compare quality, but I'd say that one of the ideas is to identify better and better metrics that approximate that comparison; otherwise most of the developments described in this forum would be pointless...

metrics != quality.

metrics like PSNR, SSIM, VQF, etc measure the difference from the source. there is some relation with quality but a higher value does not always mean higher quality since the human eye has a very complex perception that cant be just described as a mathematical function.

[Esurnir]

clip1 = DirectShowSource("C:\Foo.xxx", audio=false)
clip2 = DirectShowSource("C:\Bar.yyy", audio=false)

StackHorizontal(clip1.crop(0,0,-320,0),clip2.crop(320,0,0,0))

I personnally like this one, the difference are more striking that way in my opinion.

[LoRd_MuldeR]

My attempt of a simplistic tool to compare encodes:
http://www.mediafire.com/?bhkl0qhknne

Put it in your VirtualDub folder and make sure Avisynth is installed

[Ytterbium]

This seemed quite good, around 94% seemed good with my eyes.

http://www.ece.uwaterloo.ca/~z70wang/research/ssim/

and again in MatLab

http://www.mathworks.com/access/help.../ref/psnr.html

[Sharktooth]

read my previous post. metrics do not mean quality.
infact psychovisual enhancements usually lower the metrics but rise the perceived quality by much.
you cant automatically compare quality with a program or using metrics... you still need your eyes.

[poisondeathray]

I usually use my eyes to compare for the reasons Sharktooth gave

I've noticed a trend when taking a series of frameshots for comparison from the same frames (between 2 different encoding settings from same source, same bitrate), when converting the .bmp => .png (or even a direct .png capture), the "higher quality encoded" frame is almost always larger in size (Disc space). Of course the .bmp screenshot between the 2 encodes is the same size. I'm guessing that the "higher quality" frame has more detail retention and information in that frame, and this is an objective metric. I'm not sure what goes into the .png compression algorithm - it might be similar to the PSNR, SSIM measurements?

[Sharktooth]

no, but it can be a false measure since compression algos prefer certain data sequences instead of others...

[Ytterbium]

I agree that your eye are the best tool, It pointless in compairing 2 lossey picture files as you wouldn't be measuring the loss from codec. I think PNG is lossless, it uses compression like zip which is why png's are bigger than jpg's. bmp is certinally lossless as it's pure uncompressed data.

I found that you can do PNSR (Compare), SSIM & VqmCalc in Avisynth. SSIM is supposed to be much closer to human perception that PNSR.

[ggf31416]

Originally Posted by poisondeathray

the "higher quality encoded" frame is almost always larger in size (Disc space). Of course the .bmp screenshot between the 2 encodes is the same size. I'm guessing that the "higher quality" frame has more detail retention and information in that frame, and this is an objective metric.

by that "metric" a codec that outputs pure white noise gives the highest quality

[henryho_hk]

Considering the "performance" of common TFT LCD screen, I don't think it is a good idea to stack the video horizontally or vertically. Even the same video looks different on different position of the screen. The distorted color, brightness, contrast, etc. have a great effect on the perceived quality.