VMR9 output too bright

[LoKi128]

--- EDIT ---
Has nothing to do with x264. Look below.
--- EDIT---

Well, I got down to experimenting with x264 and I noticed that a movie I encoded seemed "washed out". It either has the brightness turned up, or the saturation turned down or both.

So I opened the AVS with MPC and took saved the frame. Then I opened the MP4 with the encode and also saved that frame. As you can see, there is an obvious difference between them.

Using x264-270, MPC v6.4.8.4, ffdshow-20050619 and Haali-20050626. Both shots taken with MPC. Commandline used to encode:

x264.exe --bframe 3 --b-pyramid --weightb --ref 3 --bitrate %2 --pass 1 --stats "%1_x264.log" --analyse all --8x8dct --subme 6 --progress --output %1_Pass1.mp4 %1.avs

AVS Used:
mpeg2source("GardenState.d2v",idct=5,info=3)
Telecide(order=1,guide=1)
Decimate()
#ColorMatrix(hints=true)
undot()

Tried with both ColorMatrix on and off, but it didnt change the output. Maybe I'm setting it up wrong and using hints mode is not enough? I remember this same question being asked in the XviD forum every once in a while, but it never happened to me with XviD.

I'm thinking that I'm missing something very obvious about this whole thing, since no one has mentioned this before. Be gentle!

PS. I'm loving x264... great quality at full DVD res at 700kbps :)

[bond]

your screenshots are not showing the same frames?

[azsd]

:) not same but similar frame....

[yokem55]

Do you have any brightness settings set with ffdshow?

[Sergejack]

similar frame ?

Have you heard about those cloud things ? ;)

You need the same frames to show the difference you're talking about.

[lexor]

You need the same frames to show the difference you're talking about.
what are you talking about? just open images in 2 windows and switch between them, the first is properly saturated and the second has white mist overlay, I can't belive you guys harras a guy about frame difference when the colour-saturation difference is so huge. And since he applies the same filtering to the entire movie and not to segments of it, filtering should either make it all white or not. Thus the problem is with his x264 encoder, either you guys can see something wrong with his command line (which I don't atm) or something is broken in the encoder.

[LoKi128]

Sorry about the two different frames! Going to the same "frame" in MPC didnt give me the same frame at all, and after a bit of moving the slider around and playing the movie, I just said "eff it" and used diff shots. Anyway, I think the diff is pretty obvious anyway.

Well, I'm back home from work, so I'll try a few different versions of x264 and see what I get. Oh and I don't have any brightness settings in ffdshow. I usually only have postprocessing and sharpen, but for this test I disabled them all.

[Pookie]

Here's my comparison of source Mpeg2 vs target x264. The x264 is a bit brighter - not too bad, however.

http://www.uploadhouse.com/images/630991343compare.jpg (http://www.uploadhouse.com/)

[LoKi128]

Well, I've been kinda busy. Anyway, played the video back with mplayer and it shows the correct brightness. So the problem has to be in either mpc or ffdshow. So i play it with WMP6.4 and it also has the correct brightness. So it is definetly MPC. I play around with the different output options:

System Default: OK
Old Renderer: OK
Overlay Mixer: OK
VMR9 (windowed): Too bright, jerky
VMR9 (renderless): Too bright

The "render video in 2D/3D" settings don't change the brightness of the output in VMR9 mode. This is all with MPC 6.4.8.4. At this point I don't know if it is a problem with MPC or with the windows VMR9 drivers or whatever.

Anyone know what could be affecting the VMR9 "side" of the output, or where to tweak that?

[LoKi128]

Did some searching, found this post (http://forum.inmatrix.com/index.php?showtopic=1090) that mentions changing the output colorspace of ffdshow to YUY2 to output the correct range of values. That didnt help. So I disabled them all and just left RGB32 (with the high quality option ticked) and that fixed the problem.

This thread is obviously not related to x264 at all. Maybe it should be moved into Software Players or some other appropriate place.

[hrlslcbr]

Hello. I also get brighter colours using VMR9, but I believe that it is not a problem of the VMR9 filter, but a problem of the overlay mixer producing colours that are too dark. The blacks of the second image you posted seemed ok to me; not too bright, but the first one is too dark.
If your source is DVD, then the colour range is 16-235 and not 0-255 so black will not (or should not) look like black on a PC monitor, but should appear slightly grey; and that's what you're getting with the VMR9 filter.
The black bars on the second image are 16,16,16 (RGB), and the ones in the first one are 0,0,0.

[Beave]

I had the same problem and discovered in some other forum that this is a typical VMR9 feature. It has the black point at around 15 % higher than it should like described in the post above. There is some advantage to this I can't remember. The same for the whites as well. You are supposed to adjust your screen so that the blacks get black again (16 should look like 0 after lowering the brightness of the display). There are even some parts of calibration DVDs that just cover this part.
My suggestion is to use overlay on normal computer monitors and VMR9 on dedicated Movie-Monitors like your HTPC under the TV.
Somewhere in the avsforum there are huge threads and how-tos about this.

[donjuan03]

My IPod 5g (video) seems to use the VRM9 type blackness and it is making my encoded videos look over bright is there any recommendations for encoding with limitations such as I have? Any help appreciated

[3ngel]

This brightness problem has been discussed in another thread "Decoder Filter for x264 decoding", and has been discovered that there is a real issue in the FFDShow decoding regarding brightness and color space.
With the default FFDShow settings the color space is junked. Allowing only RGB24 and RGB32 on FFDShow, the color space problem is gone, but the decoding remains a bit more bright than the original source (or compared to other decoders). I think atm it's a kind of a bug.

[donjuan03]

So during the encoding this problem occurs or during playback because it seems to happen during playback on my computer (unless i switch to overlay) and it happens with playback on the IPOD as well. Maybe I could encode it darker expecting the brightness gain, but I would like to find the source rather than avoid it.

[Manao]

and has been discovered that there is a real issue in the FFDShow decoding regarding brightness and color spaceIt's not quite what have been said. The issue lies within the renderer, not the decoder.I think atm it's a kind of a bug*sigh*. It isn't a bug. Do understand that the decoding process is correct. The pixel values are correct - in the boundary of dct mismatches for asp. If they are shown wrong, the issue isn't in the decoder anymore, but in the renderer. Forcing RGB output in FFDShow is a last chance workaround, but do be sure that there're other ways to achieve the same results without having to go back this early to the RGB colorspace ( which isn't a good idea, since the video card does the job much faster ).

[3ngel]

It's not quite what have been said. The issue lies within the renderer, not the decoder.If the problem would be within the renderer, the brightness matter would verify with *every* decoders decoding the content (comparing in the same VMR9 environment obvioulsy).
But indeed it happens only with FFDShow.
With VMR9 Renderless, ND (and 3vix too i've tested) decoding is faithful to the original source without the brightness altered value.

[donjuan03]

So what about the ipod where I have no control over the rendering settings, how should I accomodate to that?

[foxyshadis]

donjuan, add ColorYUV(levels="TV->PC") to the end of the avisynth script, if it is this problem. If it's just a bright backlight maybe you should just lower the movie brightness.

Since you generally need to make specialized encodings for ipod anyway, compared to PC archival versions, you might as well modify the encoding procedure. That is unless anyone knows a better way.

[cjei]

i have a similar issue with overlay.
Overlay seems to assume Rec.709 colorimetry for my 1280x720 encodes.

Are we supposed to use Rec.709 for hd encodes ??

[donjuan03]

foxyshadis, that worked thanks alot!

[mongoosesRawesome]

Forcing RGB output in FFDShow is a last chance workaround, but do be sure that there're other ways to achieve the same results without having to go back this early to the RGB colorspace ( which isn't a good idea, since the video card does the job much faster ).

Why do you say "last chance workaround"? What other ways are there? What are the drawbacks to going to the RGB colorspace?

I found no substantial increase in CPU usage when playing an x264 encoding with "high quality YV12 to RGB conversion" turned on in ffdshow and it fixed the brightness issue.

I attached some samples from an X2 encoding. The xvid shot was performed by the xvid decoder, not ffdshow. Does the xvid decoder do an RGB conversion?

Source.png (http://home.comcast.net/~falk3901/21102source.png)
X264 high quality YV12 to RGB conversion.png (http://home.comcast.net/~falk3901/21102x264RGB.png)
X264 YV12.png (http://home.comcast.net/~falk3901/21102x264YV12.png)
XVID not using ffdshow.png (http://home.comcast.net/~falk3901/21102xvid.png)

[Rash]

Was it fixed? I have the same results with both High Quality RGB convertion and no convertion.

With » http://img25.imageshack.us/img25/28/withhighqualityrgb4nq.png
Without » http://img25.imageshack.us/img25/7758/withouthighqualityrgb9hp.png

[mongoosesRawesome]

i can't see any difference in your screens, Rash.

If you look at mine though, it appears as though it works. I'm still waiting to hear why its a "last chance workaround" though.

[Rash]

Me too. ;)

[pcaca]

You can get the correct brightness if you set the input sliders in ffdshow Levels page at 16-235 and ffdshow will expand the input levels of 16-235 to output levels 0-255. Also for my taste is better if you check "Full range" in the same page.

[Visor]

add ColorYUV(levels="TV->PC") to the end of the avisynth script.
This is one of those "holy grails" that I've been looking for... for quite a while. :) I don't use this for my iPod, but I was seeing this problem when creating AVCHD discs and playing them on my Sony S350 on my Sony 52" LCD TV. Putting in that one little command keeps the brightness at the proper level for my calibrated TV.

Thanks!
Visor

[Rumbah]

The problem is simply the driver for your graphic adapter. Your card does the conversion from YV12 to RGB if the decoder outputs YV12. The results may vary from driver revision to driver revision and you simply have to try it (it was some weird reason in the VMR specification and a different RGB colour space). As suggested before setting ffdshow output to RGB is the safest thing to do. You can setup a profile so that ffdshow outputs RGB for your media player and YV12 else so you can encode your material without colour conversion.

PowerDVD for example outputs to RGB, too, to avoid these kind of problems (at least it did some time ago).