[FFmpeg-user] DPX 10bit Log in FFplay and is there a test

[Dave Rice]

Hi Jim,

> On Nov 2, 2017, at 2:38 PM, Jim Shupert <jshupert at theppsgroup.com> wrote:
> 
> Friends,
> 
> I have a couple of matters that I thought I would reach out to the communiuty for some wisdom.
> 
> 1 --------------------------------------
> FFplay showing "right" color of dpx 10bit log
> in the recent cellar ietf.org doc FFV1 and Film   by Joshua Yocum
> 
> https://drive.google.com/open?id=1-nZj53eCuY_81qYVkVDdZRa_pkHicVzETaTAxkCcYD8 
> 
> pg six reads
> 
>  Colors in ffplay -- There seems to be some inconsistency with how 10 bit log videos look in
> ffplay compared to other video players (see section on this in Detailed Comparison for more
> information).
> 
> My Q is :  what do you think?

I think the dpx decoder doesn't use the Transfer characteristic or Colorimetric specification values in a dpx header. The decoder seems to skip from the Descriptor directly to Bit depth. The dpx encoder writes Transfer characteristic and Colorimetric specification values but it is hardcoded  to:
Transfer characteristic = 2 (Linear)
Colorimetric specification = 2 (Not applicable)

There's a related ticket already at http://trac.ffmpeg.org/ticket/6023 <http://trac.ffmpeg.org/ticket/6023>.

> 2 ------------------------------------------------------
> 
> This also fits squarely in the area of wondering...
> 
> does anyone know of - or - might it be possible to
> 
> have a test of a dpx seq and Prove that it is log or linear.
> 
> presently I just eyeball f.dpx apply a lut
> 
> what I might wish i had was something like
> 
> ffprobe < test for logness > 0001.dpf

I suppose getting a histogram of the image may help in the eyeballing. For an automated approach, perhaps you could use signalstats to report the average luma value to see if it's closer overall to the linear 'middle' or the logarithmic 'middle'.

[...]

Dave Rice