I need to create a video with following conditions:

1. It can only be made from a sequence of JPEG files, each having its timestamp (milliseconds) in the name. The images’ durations are not the same, they differ, so i cannot just concat them all and use particular fps
2. There are several tar archives with the images sequences, the archives are kind of huge so I read them from a file storage as an async steam of data and cannot save them on the disk as files. The frames are read and right away put to ffmpeg running process stdin.
3. The images may have different aspect ratios so it’s required to make a NxN square and scale the images to fit in with filling the empty space with pads

My current solution:

ffmpeg -r $someFpsValue -i - -vf scale=w=$w:h=$h:force_original_aspect_ratio=1,pad=$w:$h:(((ow-iw)/2)):(((oh-ih)/2)) result.mp4

As you can see, it doesn’t let me concat the images with correct durations. I know that the concat demuxer can solve the problem of merging images with different durations but seemingly it doesn’t work with pipe protocol. I have an idea of evaluating an average fps as (videoFramesCount) / (videoDurationInSeconds) for -r argument, or maybe even counting the fps for each video’s second and then getting the avg, but maybe there is a more reliable solution (like some concat demuxer analogue)?

Thanks in advance 🙂