A dockerized version of the style-swap algorithm by rtqichen. nvidia-docker is used to make use of GPU hardware.
THIS IS WORK IN PROGRESSS
- P100: content size 1300 with decoder network
- GeForce GTX 970M: content size 450 with decoder network
- Add decoder network inside the container
- In Painnt multiscale generation seems to be taking place, with large patterns appearning in the background while keeping the structure and fine details. One way to do this might be with this approach: https://github.com/ProGamerGov/Neural-Tile/blob/master/multires.sh
- A review on style transfer methods
- Fast Patch-based Style Transfer of Arbitrary Style: paper, implementation
- Neural-tiling method