4 ms·

Kind of surprised this stuff hasn't totally blown away traditional video compression techniques yet.

• make3 8y ago
  I suspect that a combination of the two is where it's at, ie, store a lossy classical compressed version, then remove the artifacts/dream up details with deep learning
  • rasz 8y ago
    or pretrain the network on the movie you want to compress, then ship cartoon compressed version + trained network.
    • gok 8y ago
      Right that was exactly what i was thinking
• gukov 8y ago
  This got me thinking: in future we will probably stop streaming video and instead just send simple vectors. The data then gets rendered in real time into actual video. The customers will even be able to customize the movie: pick their favorite actors, the environment, etc.