Incorrect Green Levels and/or Over Saturation

[M3Rocket]

I seem to be experiencing a green saturation problem with OE 3 that might be related to the Intel video driver.

Here's a comparison:

/ / / / OpenElec HDMI output / / / / | / / / / Oppo HDMI output / / / /


------------------------------------------------------------------------------------------------------------------------------------------------


/ / / / OpenElec HDMI output / / / / | / / / / Oppo HDMI output / / / /



It does not matter whether I use software or hardware (VAAPI) acceleration. The patterns are from the AVSHD709 mp4 and Blu-Ray Calibration files/discs.

I ran the same files through an Oppo player and the Intel_X64 OE 3.0 build on a Zotac ID82. I also tried using a Windows 7 with XBMC 12.1 on an ION2 system, and the output looked like the Oppo--which is correct. So it seems like the Intel linux driver is outputing the wrong levels for green and over- or under-saturating it?

Can anyone else with an Intel please confirm what I'm getting?

[jbrady3324]

I am also seeing this as well. I haven't ran the patters on my intel linux openelec setup, but I can definitely see it during playback

I double checked the patterns via my PS3 and green is reading correctly. Definitely something up with the intel drivers.

[M3Rocket]

The more I look at this, and flipping really quickly between the two identically exposed screen shots of the incorrect Intel OpenElec/XBMC and the correct Oppo outputs, the more I think that the green AND blue output is under-saturated in this way:
Video driver green/blue channel output is likely set to Full 0-255, but actual output is 16-235 due to a video driver bug.
The movie source green/blue channel output is 16-235, but is being requantized to 0-255 resulting in the strange results because of the video driver not outputting the full range.

It is really easy to see on the green channel--the saturation of the Oppo output is much "brighter" than the Intel output.

[jbrady3324]

So I finally pulled out my meter to test the color reading. The green and blue levels are undersatured at 100% and 75%. Unfortunately, this also results two obvious issues that you can see in most material:

1. Cyan being off which a bunch of movies use today to give the movie a certain look.
2. Because blue is undersaturated, the overall picture appears to be dimmer. And as you said, if you flip back and forth between your other bluray sources you can see blue material appearing much brighter.

Your theory might be correct as it does make sense. The colors might be clipped which would cause the undersaturation we are seeing.

I might just have to recalibrate on a different TV mode for xbmc to get around this

[M3Rocket]

Nice work jbrady! I posted a link to this thread on the bugzilla issue regarding the Intel video driver output levels. Hopefully, someone from Intel will see it and look into this problem.

[jbrady3324]

Nice. Hopefully we get a fix soon. It also appears that contrast/gamma is impacted as well. But that might be a result of the green color tracking since green heavily impacts gamma tracking

[jbrady3324]

Has this been resolved?

[M3Rocket]

Nope. Still not fixed. The Intel engineer confirmed the issue but then posted something way over my head:

Ville Syrjala 2013-05-29 13:47:49 UTC

(In reply to comment #16)
I would guess that the problem is somewhere else. Maybe the CSC is using BT.601 conversion matrix instead of BT.709?

I tried to code up a small tool to generate similar patterns and can't reproduce the problem on my IVB at least. Depending on the picture settings of my TV, I can get some clipping on different channels (blue seems to clip most, then red, and green the least). In "Movie" mode I don't get any clipping however.

You should grab the framebuffer contents and see what kind of levels are actually being scanned out (assuming you're not using an overlay plane to scan out YUV directly...)

I was hoping it would get some expert's attention.

[xnappo]

So I dug around to see what I could find. The standards are described here:
http://en.wikipedia.org/wiki/Rec._601
http://en.wikipedia.org/wiki/Rec._709
As you can see - the colorspace parameters are the same except for green.

Possibly of interest?
http://forum.videohelp.com/threads/32986...ost2045830

[EDIT] Further reading suggests that if the tags are not set properly in the test file, then the wrong matrix may be getting used.

xnappo

[lmyllari]

I ran into this same issue on the Haswell NUC. I've been reporting my findings over at the xbmc forum.

It looks like the issue is not with the framebuffer output. Using software decoding (either with another player like mplayer2, or xbmc) fixes the issue for me.

I looked into the opengl colorspace conversion, and it seems fine. I think the problem is with VAAPI decoder output (which sounds strange as h264 decoding should be bit exact and thus probably verified at Intel) or with the renderer (which looks pretty straightforward).

Next step is to try with VA-API enabled mplayer or gstreamer and see if they show the same issue.

edit: I can reproduce the issue with mplayer-vaapi