Our model are based on pipeline of two Enhanced Deep Residual Networks for Single Image Super-Resolution https://arxiv.org/abs/1707.02921 networks:
- 32*32 -> 64 * 64 - network 20m parameters
- 6464 -> 256256 - network 10m parameters
For 3232 -> 256256 network trained seperatly and then united and trained.
Example on Validation set:
