Open runvnc opened 7 years ago
I have seen some papers about Image Compression with DNN. Also there is this repo with a guy with some results (but with a very simple network).
The problem I see is that you need a lot of different videos computational power. I think you will get better results with Recurrent Neural Networks that is somehow the way video codecs work, but you end up with a huge network.
I keep hearing about things like CNNs winning at image recognition contests etc. and for awhile have assumed that eventually a new video codec that involves distributing a deep network trained on many videos would achieve breakthrough compression results.