Tips on crawling images from video

cvondrick / soundnet

SoundNet: Learning Sound Representations from Unlabeled Video. NIPS 2016

http://projects.csail.mit.edu/soundnet/

MIT License

462 stars 94 forks source link

Tips on crawling images from video #7

Closed keunwoochoi closed 7 years ago

keunwoochoi commented 7 years ago

Hi, I'd like to ask some tips that would be generally applicable in video/image stuff deep learning, I've been only working on music-related works. Some (if not all) might seem dumb ;)

What would be good image format of extracted frames of video? jpeg or png?
What was image sampling rate in the work? -- how many images per second did you sample?
Any other tip/hack would be appreciated.

Thanks!

cvondrick commented 7 years ago

I would recommend JPEG.
We extracted 1 frame per second.
Video is huge, so I recommend designing efficient ways of storing them!