[FFmpeg-devel] [PATCH] libavfilter: temporarily remove DNN framework and vf_sr filter

[Pedro Arthur]

2018-07-28 14:31 GMT-03:00 Pedro Arthur <bygrandao at gmail.com>:
> 2018-07-27 14:24 GMT-03:00 Rostislav Pehlivanov <atomnuker at gmail.com>:
>> On 27 July 2018 at 18:12, Rostislav Pehlivanov <atomnuker at gmail.com> wrote:
>>
>>>
>>> And the coding style is just the tip, there are dozens of things I
>>> disagree with, such as the custom float to 8 bit conversions, when filters
>>> already exist which take in floats (zscale).
>>>
>>
>> I should probably expand on what I mean by the custom float conversion:
>> make the filter accept a float pixel format, such as AV_PIX_FMT_GBRPF32. We
>> have one, we use it in vf_tonemap and vf_zscale. Users will need to specify
>> a filter to convert to float first, which will match what they have to do
>> for vf_zscale and vf_tonemap. If you need to process YUV then add a float
>> YUV format, but please, don't do custom conversions in filters.
> It seems there is a patch adding YUV float formats in the ml, I asked
> the student to make a patch removing the conversion and supporting
> only float formats when the yuv float format get in the tree.
Just to update on the matter, there are currently two patches on ml,
one which fixes the code style and one which removes a huge amount of
stored data from the sr filter.
We are working on removing the format conversion from sr filter,
adding the needed floating formats to swscale.