i am encountering a little problem with a video-data and I was wondering if the fastai library could help me out with it. Basically I am trying to predict the speed of a car using a video (about 17mins long), I already segmented it into 20k-ish frames using OpenCv, but haven’t thrown the frames yet in a DL model. Any suggestions on how I should approach the problem? Is the Fastai Library suitable for video data (although now they’re just images)?
If you have a “short” video and you have enough space/time/patience to convert it in images, then loading them into a Fast.ai DataBunch is a breeze (watch Lesson 1, it’s pretty straightforward).
Now I’d like to raise the scale at which I can make Fast.ai process video data because extracting images simply doesn’t scale in any of the above dimensions, specially patience, and so I was looking for a VideoDataBunch or VideoDataLoader that, unfortunately, dont’ exist.
I’ve found this thread that gives a starting point: Video processing and augmentation
But this problem shouldn’t be so uncommon, there should be something ready to use…