Image Classification and Convolutional Neural Networks (CNNs)

Overview

This project explores image classification using Convolutional Neural Networks (CNNs) in PyTorch. It involves working with a subset of the SUN397 dataset, performing image preprocessing, implementing gradient-based optimization, and training a CNN for scene classification.

Features

• Exploring the SUN Dataset

  • Work with a subset of the SUN397 dataset (20 categories, 50 images per category).
  • Use PyTorch’s DataLoader for image loading and manipulation.
  • Analyze dataset statistics (e.g., counting portrait-mode images per category).

• Curve Fitting with PyTorch

  • Fit a polynomial function to given data points by optimizing four parameters.
  • Use PyTorch autograd for automatic differentiation.
  • Plot predictions against the original data.

• Training a Convolutional Neural Network (CNN)

  • Implement a CNN model for scene classification.
  • Set up dataset preparation, hyperparameters (learning rate, batch size, optimizer, epochs), and train the model.
  • Evaluate performance using validation accuracy.

• Making Predictions with the Trained CNN

  • Process input images and obtain softmax probabilities for different scene categories.
  • Display and interpret the model’s predictions.

Technologies Used

• Python
• PyTorch
• NumPy
• Matplotlib

Acknowledgments

This project is based on CS 1674/2074: Image Classification and CNNs coursework. The dataset is derived from the SUN397 dataset.