U net semantic segmentation

Implementation of the U-Net for semantic segmentation

Colab link

• Implemented the U-Net
• The architecture of U-Net is as below

• The model is trained on the cityscape dataset.
• batch_size = 16
• epochs = 10
• lr = 0.01
• Adam optimizer
• CrossEntropyLoss()

Step loss and epoch losses

Implemented the Gaussian process in the middle to accommodate the temporal camera pose data