Object Classification with PyTorch

This project implements an image classification pipeline using PyTorch and a ResNet-18 backbone on a custom dataset with 8 classes. Data augmentation techniques are applied to improve model generalization.

Objective

Train a deep learning model to classify images into 8 categories while ensuring model interpretability and robustness.

Tools & Technologies

• Python, PyTorch
• Jupyter Notebooks
• Data augmentation: random flips, rotations, affine transforms, color jitter, Gaussian blur, noise, random erasing
• Grad-CAM for model interpretability

Training & Evaluation

• Training: Cross-entropy loss, Adam optimizer, learning rate scheduling, early stopping
• Validation: Monitor accuracy and loss on validation split
• Checkpointing: Save best model weights
• Testing: Evaluate on test set and export predictions to CSV

Model Interpretability

• Grad-CAM visualizations highlight image regions most influencing model decisions, supporting explainable AI.

Key Takeaway

This project demonstrates practical skills in deep learning, model training, evaluation, and interpretability techniques, applied to real-world image classification problems.