Original Pix2Pix paper: https://phillipi.github.io/pix2pix/
Generator is a U-Net like network, the architecture is demonstrated in the picture:
Data downloading script scripts/download_data.sh supports four classical Pix2Pix datasets: facades, maps, edges2shoes and night2day. There is also an additional dataset churches: download link. I have collected this dataset on my own. The dataset is organized in accordance with classical Pix2Pix datasets. Each pair contains a photo of a building itself and the edges of the photo. There are 250 training and 50 testing pairs.
Running main.py starts training with the configuration stored in config.py file
python main.pyThe training process can be observed in these gifs:
