No-Reference-Image-Quality-Assessment-using-BRISQUE-Model VS media2

They originally wanted the system mainly for video quality assessment, but were far more interested in it for several other things once we started showing it off. I did have some success modifying Krshrimali's Brisque image quality assessment code into a library with an API which I could process individual video frames with, but this turned out to be very slow without GPU acceleration (Neighborhood of 2 seconds per frame IIRC for 1080, IIRC,) and didn't put a lot more experimentation into it. I think even with GPU acceleration it'd be hard to get that down to real-time. Might be OK for short clips, though.

The ffmpeg command line is a shitty programming language. I'd rather work with it in some other shitty programming language.

I use it in conjunction with the moldy old ffmpeg C API for video processing. Here's the current iteration of my wrapper library -- I'm still missing something in remuxing, but it works pretty well. I haven't done an update to it in a while because work's been busy. I've been doing a bunch of cloud stuff there and am meditating on rolling a new iteration of this library with cloud functionality baked in. The segmenter/desegmenter stuff in the current library works pretty well for that sort of thing, but it feels a bit awkward to use.