Low bandwidth video experiment using dlib to implement face-landmarking and an auto-encoder.
One hackathon at Microsoft I tried to invent a super low bandwidth codec that would be good for 2G networks. It got a lot of attention in the hackathon, but nothing came of it. Part of the idea was to use face-landmarking to highlight expressions of the user. This was useful when the face was blurry.
This was the initial version before I implemented an auto-encoder based codec. I will add that code when I can simplify dependences. This version includes the version of dlib I used at the time.