Digging in some deep learning for visual transformation inspired by my dearest friend/artist/educator Gene Kogan's Alice In Wonderland and the amazing deepslowflow.
The performance requires a high-end GPU system with a recommended (must require or you will wait till you die) Nvidia + CUDA6.5+ backend and optimizer. Since macbooks have replaced all Nvidia Graphic cards to AMD (dunno why), I had to perform the renderings on AWS EC2 gpu servers, until either I get a new gaming computer, or I get tired of this thing.
Some results for images, see more videos on Instagram
Video of the collection
Some customized results for videos (GIF version) with optical flows
15 styles of Shanghai
6 styles of Chewy
References:
Code: Justin Johnson
Instruction: Gene Kogan
Instruction: AWS tutorials
Tips: Torch7 CheatSheet
Code: DeepFlow
Instruction: ffmeg Documentation for videos