Code: github.com/PSRahul/semanticseg
A U-Net based neural network was trained from scratch using PyTorch Lightning on top of PyTorch. The dropout probability was varied to optimise the network.
Dataset
Oxford-IIIT Pet Dataset — O. M. Parkhi, A. Vedaldi, A. Zisserman, C. V. Jawahar, “Cats and Dogs”, IEEE Conference on Computer Vision and Pattern Recognition.
Training setup
- Optimiser: Adam with the default learning rate of .
- Loss: cross-entropy between predicted pixel labels and ground truth.
- Callbacks: early stopping and best-validation-loss checkpoints.
Results
