Paco's Puppy Problem

Overview

What I built

This project focused on building and evaluating image-classification pipelines in PyTorch. I implemented a CNN and a Vision Transformer, explored transfer learning with different frozen-layer configurations, and designed a deeper regularized CNN using batch normalization, dropout, and weight decay. The experiments showed that transfer learning improved held-out generalization and that the CNN outperformed the ViT on this limited-data task.