Semantic Segmentation of Pets

GitHub Link

A U-Net based neural network was trained from scratch using Pytorch Lightning wrapper over the Pytorch Framework.

The dropout probability was varied to optimise the network.

Dataset

https://www.robots.ox.ac.uk/~vgg/data/pets/

O. M. Parkhi, A. Vedaldi, A. Zisserman, C. V. Jawahar
Cats and Dogs
IEEE Conference on Computer Vision and Pattern Recognition

Model

A U-Net based neural network was trained from scratch using Pytorch Lightning wrapper over the Pytorch Framework.

The dropout probability was varied to optimise the network.

Optimiser

Adam with the default learning rate of 1^-3.

Loss

Cross Entropy Loss of classified pixel labels and ground data.

Callbacks

Early Stopping , Best Validation loss checkpoints.

Results