[FFmpeg-devel] [PATCH v4][GSOC] avfilter: added colorconstancy

[Thilo Borgmann]

Am 24.07.18 um 19:07 schrieb Paul B Mahol:
> On 7/24/18, Rostislav Pehlivanov <atomnuker at gmail.com> wrote:
>> On 24 July 2018 at 17:10, Paul B Mahol <onemda at gmail.com> wrote:
>>
>>> On 7/24/18, Rostislav Pehlivanov <atomnuker at gmail.com> wrote:
>>>> On 23 July 2018 at 21:36, Thilo Borgmann <thilo.borgmann at mail.de> wrote:
>>>>
>>>>> Am 19.07.18 um 08:26 schrieb Thilo Borgmann:
>>>>>>
>>>>>>
>>>>>>> Am 16.07.2018 um 13:41 schrieb Mina <minas.gorgy at gmail.com>:
>>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>>   This patch introduces Grey-Edge algorithm as part of the Color
>>>>> Constancy Filter project in GSOC.
>>>>>>>
>>>>>>> V4 changes:
>>>>>>> - Fixed error in filter.texi that resulted in breaking "make
>>>>> doc/ffprobe-all.html"
>>>>>>
>>>>>> If there are no more comments coming in I`ll push this in a couple of
>>>>> days.
>>>>>
>>>>> Applied.
>>>>>
>>>>> -Thilo
>>>>> _______________________________________________
>>>>> ffmpeg-devel mailing list
>>>>> ffmpeg-devel at ffmpeg.org
>>>>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>>>>
>>>>
>>>> What's the point of having an imperfect color conversion filter that
>>> could
>>>> screw up compared to a mathematically sound approach?
>>>> After realizing what this filter does I'm kinda against it and would
>>> rather
>>>> we not have an unmaintainable, unused color conversion filter and I
>>>> think
>>>> if there's no reasoning behind it I'd like it reverted.
>>>> _______________________________________________
>>>> ffmpeg-devel mailing list
>>>> ffmpeg-devel at ffmpeg.org
>>>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>>>
>>>
>>> This is not color conversion filter.
>>> _______________________________________________
>>> ffmpeg-devel mailing list
>>> ffmpeg-devel at ffmpeg.org
>>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>>
>>
>> Use whatever terminology you'd like, its still not something I'd consider
>> acceptable. I want to be able to reproduce the results and the NN weights
>> and I can't.
> 
> This does not use NN weights at all. You are in wrong thread.


It is pixel stats only and yes, I guess Paul is right you've just missed the actual thread you're complaining about.

However, Mina plans to possibly reuse some of the NN code that comes with the other GSoC project and therefore let's find a solution there to get a sane solution to the problem.

-Thilo