anti-epilepsy bright flash remover video filter

[compn]

any chance a filter to remove bright flashes from video?

it may drop the frame or duplicate a previous non-bright frame, making video stutter a bit.

i'll upload a sample soon.

[Carl Eugen Hoyos]

Could you provide a sample?

[compn]

uploaded to incoming as bright-flash-sample-for-filter.mkv
4mb.

[Elon Musk]

please always put link here

[Carl Eugen Hoyos]

Sample from ​http://en.wikipedia.org/wiki/Denn%C5%8D_Senshi_Porygon

[compn]

this sample is also good for testing:
​https://www.youtube.com/watch?v=qKSu_GdF6wk

[Vladimir Panteleev]

Hi,

I've been working on a filter to do this. My code so far is here:

​https://github.com/FFmpeg/FFmpeg/compare/master...CyberShadow:epilepsy

Feedback from those who would find this useful would be appreciated.
I can also provide Win64 ffmpeg/mpv binaries.

I intend to submit the filter for inclusion once we conclude the results to be satisfactory.

[Marton Balint]

Really nice, some comments at first glace:

• Use the .activate callback
• You can probably factorize the blending code in vf_framerate so ASM can be used for blending
• A separate mode to detect only flashes (and not remove them) would probably be useful

[Cigaes]

This is bug tracker, patches should be discussed on the mailing list.

Also: https://ffmpeg.org/pipermail/ffmpeg-devel/2019-February/240556.html

[compn]

yes if you can provide a win64 binary I can test, although I am no expert.

I am curious on why you chose to write a filter for this. I am happy to see work on this, I hope a filter like this can help people.

I am assuming it is real time ?

[Vladimir Panteleev]

Replying to compn:

yes if you can provide a win64 binary I can test, although I am no expert.

There are binaries available at this very-temporary URL:

​http://89.28.117.31:8081/bin/

There are also some samples/comparisons on the same server, but note that they may not be safe to look at.

I am curious on why you chose to write a filter for this.

Someone I know mentioned that they would find such a thing useful, and it seemed like a very interesting project that could help people.

I am assuming it is real time ?

Yes, it works in real time (on recent CPUs) and does not buffer.

[compn]

I like your comparison videos. Very clear before and after.

Your filter tries pretty hard on this clip from the incredibles 2, but I don't think theres anything that can be done for it.

​https://www.youtube.com/watch?v=8L_9hXnUzRk

the only other thing i could think of is that the "next gen" version of this filter would remember which macroblocks are moving/changing colors at too fast of a pace, and then you could just freeze those ones while the (still moving, but not flashing) macroblocks are allowed to continue. Not even sure if its possible to do what I am describing though. It would also cause or look like tearing.

is there a memory leak? seems to eat about 3gb+ ram here when using your filter vs no filters. could be a mingw bug or something else.

but yes, if you have not already, please submit patch to ffmpeg-devel for discussions. that is the proper place for patch discussions. this bug tracker does not get as many eyeballs as the mailing list.

[Elon Musk]

Fixed in c888adf590eeaf8f854de9067886ecb2f0f08419.

[compn]

I asked netflix about the filter. they said it would need proof that the filter is following the ITU standard before considering using it.

​https://www.itu.int/dms_pubrec/itu-r/rec/bt/R-REC-BT.1702-2-201910-I!!PDF-E.pdf

maybe a future feature request. just putting itu link here for now. maybe someone would put a bounty or hire developer to make it conform to specs.