This a simple and fun little classifier I made, which can classify between bell bottoms and ripped jeans. Additionally, I used the "hooks" functionality of PyTorch to generate Grad-CAM heatmaps and overlaid it on top of the original image. This gave me the opportunity to understand how the classifier works under the hood.
The notebook explains the steps and process in detail, so that it is easy to follow along.
We see how when making the prediction, the model focuses on the knees and the trouser leg area. This is pretty intuitive since bell bottoms have their signature bell shape in the trouser leg area, and jeans are mostly ripped near the knee region.