Mask-ShadowNet: Toward Shadow Removal via Masked Adaptive Instance Normalization

This is the Pytorch implementation of "Mask-ShadowNet: Toward Shadow Removal via Masked Adaptive Instance Normalization".

Note: The resolution of test shadow image should be 256*256 for our pre-trained model.

Please first set the dataset path then start training or testing.

• Training

cd script
bash train.sh gpu_id display_port

where gpu_id is the ID of gpu, and display_port is the port number of visdom server.

• Testing

cd script
bash test.sh gpu_id

Prereuisites

• Linux or macOS
• python 3
• CPU or NVIDIA GPU + CUDA CuDNN
• PyTorch 1.2+