In the Assignment 1 folder, there is a notebook whose objective is to develop a computer vision system that, using only traditional computer vision techniques, is able to identify certain books from a single image of a bookshelf, given their reference images.
In the Assignment 2 folder, there is a notebook whose objective is to implement a NN that classifies images (with pythorch) of 37 breeds of cats and dogs from the Oxford-IIIT-Pet dataset.
The assignment is divided into four parts:
- implementation from scratch of our own neural network,
- incorporates selected elements from VGG, Inception, and ResNet
- 75% test accuracy and estimated Total Size = 112.96MB
- ablation study,
- fine-tune ResNet-18,
- error analysis.