ASAP_pytorch_implementation

This is a casual pytorch implementation of spatially-adaptive pixelwise network for fast image translation

I tested the code in database of CMP Facade Database

I provided a sample checkpoint which generate the following result example.

A result sample

Image caption: Label masks, results, and ground truth image

Still some issues left to solve:

1. In a single NVIDIA GenForce GTX 1080ti, the maximum image size can only be set to 256 x 256
   
2. Some artifacts remain in the right conner that may disappear if keep on training.