On Wed, Jul 29, 2020 at 9:12 AM Jaeho Bang <jaeho.b...@gmail.com> wrote: > > Dear members of ffmpeg-devel, > > I am currently working on a research project and from looking over the > examples and documentation in the ffmpeg repository, I believe I need to > implement a custom encoder / decoder to perform my task. > > To first explain the project idea, currently I am working mainly with > surveillance footages. Using a traditional video decoder, I convert the > video into series of image frames and perform clustering to create frame > groups. Then using some performance metric, my algorithm selects a > representative frame from each group. By implementing a custom encoder and > decoder, I aim to improve the speed of extracting these representative > frames. > > To further explain, I would like the custom encoder to be able to force the > frames I have chosen (representative frames) to be i-frames. The assumption > is that when analyzing videos, we only need to evaluate these > representative frames instead of the whole video (this I have already > experimentally confirmed), and extracting i-frames is faster than > extracting a combination of i-frames, b-frames, and p-frames. > > Also, I would like to implement a decoder to quickly extract the series of > frames I give as input (they will all be i-frames). Through preliminary > experiments, I realized that extracting only i-frames by using the command > 'select pict_type\I' query command only gives a 2x speed boost compared to > extracting all frames within the video. Hopefully, by implementing a custom > decoder, I aim to further increase this extraction speed. > > Is there any code I can refer to? Is my assumption of 'if I were to > implement a custom decoder, the extraction speed would be faster than using > the keyword select to get all the i frames' valid? > > Any help would be greatly appreciated. Thank you and have a wonderful day. > > Sincerely, > Jaeho Bang > > -- > *Jaeho Bang* > *PhD Student in Computer Science* > Georgia Institute of Technology > North Ave NW, Atlanta, GA 30332, USA
IMO, the x2 speed boost is a positive result, depending on the GOP size. Decoding a series of P frames is much faster than for the same amount of I frames. I don't believe some 'custom decoder' can improve the situation further. BR, Alex Cohn _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
