4 ms·
This but for video using the "infilling" version for changing parts between frames. The structural changes per frame matter much less. Send a 5kB image every k
by nl 4y ago
This but for video using the "infilling" version for changing parts between frames.
The structural changes per frame matter much less. Send a 5kB image every keyframe then bytes per subsequent image with a sketch of the changes and where to mask them on the frame.
Modern video codecs are pretty amazing though, so not sure how it would compare in frame size
- willbudd 4y agoI've been thinking about more or less the same idea, but the computational edge inference costs probably makes it impractical for most of today's client devices. I see a lot of potential in this direction in the near future though.
- nl 4y agoI think it's unclear how much computational resources the uncompression steps take. At the moment it's fairly fast, but RAM hungry. But this article makes it clear that quantizing the representation works well (at least for the VAE). It's possible quantized models could also do decent jobs.