Pytorch implementation of U-Net: Convolutional Networks for Biomedical Image Segmentation for segmentation of aerial maps into google maps.

I used the pix2pix maps dataset available over here

• pytorch: 0.4.1
• torchvision: 0.2.1

Generated Label Input

P.S: It took 12 hours to train on a 1050Ti with a batch size of 5 for 100 epochs. If I tried to increase the batch size, I ran out of memory. I asked around and people suggested using checkpoint and I found a discussion post related to it. Haven't tried it yet, therefore any suggestion or crtiques are always welcome.