This project tries to solve a multi-class semantic segmentation problem using UNET on the Cityscape Image Pairs dataset. This project implements a variation of the orignal UNET by adding batch normalisation between layers and using same padding.
The inputs to the model is an RGB image of a cityscape.
The output is a semtantically segmented mask where different objects are color coded.
A .pth.tar file can be found here containing weights of this model trained for 110 epochs.
This was a personal project worked on by: