[FFmpeg-user] ffmpeg makes some sources darker

[Jim Worrall]

> On 2016 Jan 4, at 3:29 PM, Andy Furniss <adf.lists at gmail.com> wrote:
> I think it's because the master is flagged as rec709 but the yuv and I
> guess hevc aren’t.

Although my opinion is not worth much, I suspect you are onto it here.
> 
> Some players (not mplayer) would assume 709 for >= 720p so you wouldn't
> notice with them.
> 
> It seems ffmpeg swscale/png honors the 709 for the master.
> 
> testing -
> 
> ffmpeg -i in.mkv -vframes 1  frame1-master.png
> 
> ffmpeg -i in.mkv -vframes 1  -vcodec rawvideo -f rawvideo frame1-master.yuv
> 
> ffmpeg -f rawvideo -pix_fmt yuv420p -s 1920x1080 -i frame1-master.yuv frame1-yuv.png
> 
> and frame1-yuv.png is darker than frame1-master.png
> 
> do
> 
> ffmpeg -f rawvideo -pix_fmt yuv420p -s 1920x1080 -i frame1-master.yuv -vf scale=in_color_matrix=bt709  frame1-yuv-709.png
> 
> and frame1-yuv-709.png visually (but not md5sum) matches frame1-master.png

I’m not sure if you’ve already done this, or you’re predicting.  
In any case, you’re not going to like the result.   
frame1-yuv-709.png is actually intermediate in darkness to the other two,
but closer to frame1-yuv.png.

However, I should add that the x265 guy, Ma0, went ahead and transcoded my minute of video
with what I think would be this, and got no difference in darkness:

ffmpeg -i min_source.mkv -f yuv4mpegpipe - | x265 --y4m - --preset placebo --subme 7 --merange 121 --deblock -1 --rdoq-level 1 --colormatrix bt709 --bitrate 3000 --pass 1 -o slow.hevc && \
ffmpeg -i min_source.mkv -f yuv4mpegpipe - | x265 --y4m - --preset placebo --subme 7 --merange 121 --deblock -1 --rdoq-level 1 --colormatrix bt709 --bitrate 3000 --pass 2 -o slow.hevc

Aside from the insane settings (my computer is doing this at 0.1 fps!!), note the --colormatrix bt709 - 
I’m guessing that was the key.  Not sure why it comes out intermediate in your test - probably
because it is already darkened during the conversion to yuv, but not in the above?

I will test this further when my processor allows.  In the meantime, thanks
very much to Andy and Carl for taking the time to look at this.