Odtnhj

I'm compiling a dataset of images for a machine learning project, and the easiest way to collect thousands of relevant images is by extracting frames from video.

However to save on time, and given slow bandwidth, it would be most efficient if I could download segments of a video, and process them that way.

I can do this easily on the command line via

curl "url" -o "path/to/outputfile.mp4" -r "startbyte-endbyte"

And I can normally split a video file into individual jpeg images via

ffmpeg -i "path/to/video_file.mp4" "path/to/output/image%06d.jpg"

However, the problem I have run into is that ffmpeg does not like it when I give it a video fragment, and gives the following error

Invalid data found when processing input

Further, attempting to use the tool Untrunc gives the following error

Failed to parse atoms in truncated file

I'm looking for any way to be able to process these file fragments. I technically have access to the entire file, and can download any part of it as needed (i.e. meta data in the beginning?), I just want to avoid downloading the whole thing.

In which direction should I go? I don't have any understanding of how video files work, but am willing to get my hands dirty if need be.

In summary, I need to get a randomly selected fragment of a video to be capable of being processed (i.e. have frames extracted).

ffmpeg should support byte range requests. So you should be able to simple do something list this:ffmpeg -ss 1800 -i http://myserver/file.mp4 -vframes 1 -an -y %d.jpg