[FFmpeg-devel] [PATCH v1 1/2] lavc/vp9: set yuvj pixel format for full range decode

[Leo Izen]

On 6/19/23 04:14, Thilo Borgmann wrote:
> Am 18.06.23 um 23:21 schrieb Leo Izen:
>> On 6/17/23 10:26, Thilo Borgmann wrote:
>>> Am 17.06.23 um 16:02 schrieb Leo Izen:
>>>> On 6/17/23 04:11, Thilo Borgmann wrote:
>>>>> While the yuvj pixel formats are deprecated lots of code still relies
>>>>> on them to be set. Without setting a yuvj420p pixel format VP9
>>>>> decoding ends up incorrectly due to auto conversion.
>>>>>
>>>>
>>>> I oppose this on principle. If there's code that relies on YUVJ 
>>>> being set, then *that code* needs to be changed so it respects the 
>>>> AVFrame->color_range field. Which code is working improperly with this?
>>>
>>> I don't like adding YUVJ stuff either. If I do
>>>
>>>   ./ffmpeg -i full-range-in.mp4 -c:v libvpx-vp9 -lossless 1 
>>> lossless-out.mp4
>>>
>>> and then comparing the frames, they are not equal. E.g. by
>>>
>>>   ./ffmpeg -i full-range-in.mp4 -i lossless-out.mp4 -filter_complex 
>>> ssim -f crc -
>>>
>>> they are not 1.0 in ssim terms.
>>>
>>
>> Are you sure? I just tested a sample and found that I got exactly 1.0 
>> in ssim terms. Do you have a link to a sample for which this fails?
> 
> IIRC I had the same impression when testing, I think I mixed up patched 
> and unpatched ffmpeg builds.
> 
> Anyway, happy you retest, I used
> 
> ./ffmpeg -f lavfi -i testsrc=duration=10:size=1280x720:rate=30 -pix_fmt 
> yuvj420p -color_range pc full-range-in.mp4
> 
> to generate my input sample. I cannot test myself again until Thursday, 
> my Laptop is not equipped with libvpx.
> 

I used: ./ffmpeg -i input.mkv -vf libplacebo=format=yuv420p:range=pc -c 
ffv1 full-range-in.mkv

I wonder if using yuv420p with pc range changes the results. Running 
ffmpeg -i full-range-in.mkv by itself reports ffv1, yuv420p(pc, 
progressive) as its format.

- Leo Izen