[FFmpeg-devel] [PATCH] avfilter/vf_chromakey: The chromakey filter preserves non-opaque alpha transparency.

[Paul B Mahol]

On 6/12/20, Gavin Smith <gavin.smith at playerbites.com> wrote:
>
> On 12/06/2020 16:58, Gavin Smith wrote:
>> On 12/06/2020 12:59, Timo Rothenpieler wrote:
>>> On 12.06.2020 00:31, Gavin Smith wrote:
>>>>
>>>> On 11/06/2020 22:03, Timo Rothenpieler wrote:
>>>>> On 11.06.2020 22:27, Gavin Smith wrote:
>>>>>> This is to address trac ticket #8724. The filter did not preserve
>>>>>> alpha transparency. Items that were transparent in the input would
>>>>>> appear black on the output or pixels that were semi-tranparent
>>>>>> would appear opaque.
>>>>>>
>>>>>
>>>>> What is the performance impact for inputs without an alpha channel?
>>>>
>>>> Firstly, I'm new to this world of filters.  Now my patch only
>>>> applies to chromakey and not chromahold. On that note, is it not the
>>>> case that the chromakey mandates that all its inputs surfaces
>>>> contain alpha (as per query_formats function)? Would any
>>>> AVPixelFormat that does not match query_formats get converted to a
>>>> format containing alpha transparency?
>>>
>>> But it still adds new code, and given it's in the very hot path of
>>> the filter, it can easily add a performance penalty.
>> Sure. I understand. I'll see what numbers I can get.  Having a quick
>> cursory look over the code, I wonder if my code actually might be
>> slightly more performant because it checks the case if a0 != 0.
>> However, this jmp/cmp may negatively impact performance.
>
> vf_chromakey wll now have 2 implementations of chromakey depending on
> whether "use_alpha" is selected. What are your feelings on the best way
> to approach this?
>
> 1. Add an additional function pointer to ChromakeyContext: In
> do_chromakey_slice(..), call the said func ptr in the loop each iteration?

This is best solution.
You could also add function pointer for do_chromakey_pixel(), which
would pass in 2nd argument current alpha value.
In one case that 2nd argument would not be used and in different case
it will be used.

>
> 2. Add a conditional to do_chromakey_slice(...): In the loop iteration,
> check "use_alpha" variable and execute and jmp to pertinent code?

That would be slow.

>
> 3. Any other suggestions?
>
> Thanks.
>
>>>
>>>>>
>>>>> Generally looks fine to me, but might need hidden behind an option,
>>>>> as to not break existing setups that rely on this filter
>>>>> discarding/ignoring the input alpha channel.
>>>>
>>>> Yes. No problem. "preserve_transparency" and default to the
>>>> equivalent of 'false'?
>>>
>>> That seems a bit clunky. Maybe something like "use_alpha"?
>> Sure.
>>>
>>> _______________________________________________
>>> ffmpeg-devel mailing list
>>> ffmpeg-devel at ffmpeg.org
>>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>>
>>> To unsubscribe, visit link above, or email
>>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>
>> To unsubscribe, visit link above, or email
>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".