Assignment 1: Image Classification

Table of Contents

1. Feature Extraction + k-NN
   • Implementation
   • Results
2. Transfer Learning with ResNet18
   • Implementation
   • Results
3. Neural Network from Scratch
   • Implementation
   • Results
4. Comparison Between Models
5. Observations and Challenges

Feature Extraction + k-NN

Implementation

• Implemented feature extraction using a pre-trained ResNet-18 model.
• Data preprocessing, transformation, and loading.
• Feature extraction from the last layer.
• Saving features for k-NN classification.

Results

• Detailed results of k-NN classification.
• Example images and their classifications.

Transfer Learning with ResNet18

Implementation

• Preprocessed and normalized images using ImageNet mean and standard deviation.
• Split the training data into train and validation sets.
• Loaded pre-trained ResNet-18 model.
• Fine-tuned the model with a custom classification layer.
• Trained with different optimizers and learning rates.

Results

• Test accuracy on test data.
• Validation accuracy during training.
• Classification report and observations.

Neural Network from Scratch

Implementation

• Preprocessed and normalized images using training data mean and standard deviation.
• Split the training data into train and validation sets.
• Designed a custom convolutional neural network (CNN) from scratch.
• Implemented data loading and training with cross-entropy loss.
• Explored different CNN architectures.

Results

• Test accuracy on test data.
• Validation accuracy during training.
• Classification report and observations.

Comparison Between Models

• Compared k-NN, fine-tuning, and custom CNN approaches.
• Analyzed differences in accuracy and computational requirements.
• Discuss the impact of hyperparameters.

Observations and Challenges

• Summarized observations and challenges faced during the assignment.
• Discussed key learnings from each approach.

---

This README provides a structured format for your GitHub repository, making it easier for users to understand the contents of your image classification assignment and access relevant information and results.