Dogs vs. Cats Image Classification

Goal: Build a binary image classifier with Transfer Learning to distinguish between dog and cat images.

Background

This dataset was originally released as part of a Kaggle competition hosted by Microsoft Research in 2013. It became a classic benchmark for beginner and intermediate deep learning practitioners due to its simplicity and effectiveness in demonstrating convolutional neural networks (CNNs) in image classification tasks.

Highlights

• Implemented using PyTorch and Torchvision
• Transfer Learning with pre-trained ResNet-50 weights
• Used Cross Entropy Loss and evaluated with F1 Score
• Achieved 98% weighted F1 Score in test set

Processing & Treatments

• Images were resized and normalized using standard ImageNet preprocessing
• Data was split into training and validation sets with stratification
• Applied basic data augmentation (e.g., random flips and crops) to increase robustness