This project implements a deep learning model for image classification on the CIFAR-10 dataset using ResNet18 architecture. CIFAR-10 contains 60,000 32x32 color images across 10 object categories (airplanes, automobiles, birds, cats, deer, dogs, frogs, horses, ships, and trucks).

• ResNet18 Implementation: Built from scratch with residual connections to address vanishing gradient problem
• Baseline Comparison: Simple CNN baseline for performance comparison
• Data Augmentation: Random cropping and horizontal flipping for improved generalization
• Comprehensive Evaluation: Accuracy metrics, learning curves, confusion matrices, and misclassified examples
• Detailed Analysis: Side-by-side comparison of ResNet18 vs simple CNN

Install the required packages using:

pip install -r requirements.txt

• PyTorch >= 2.0.0
• torchvision >= 0.15.0
• numpy >= 1.24.0
• matplotlib >= 3.7.0
• seaborn >= 0.12.0
• scikit-learn >= 1.3.0

1. Open the notebook in VS Code:

   • Open VS Code
   • Install Jupyter extension if not already installed
   • Open C10.ipynb

2. Run the notebook:

   • Execute cells sequentially from top to bottom
   • The dataset will be automatically downloaded on first run
   • Training will take some time depending on your hardware (GPU recommended)

3. Expected Runtime:

   • With GPU: ~20-30 minutes for 50 epochs
   • With CPU: ~2-3 hours for 50 epochs

Project/
├── C10.ipynb              # Main Jupyter notebook
├── requirements.txt       # Python dependencies
├── README.md             # This file
└── data/                 # CIFAR-10 dataset (auto-downloaded)

1. Setup and Imports: Import necessary libraries and configure environment
2. Data Loading: Load CIFAR-10 with preprocessing and augmentation
3. Model Architectures:
   • ResNet18 implementation with residual blocks
   • Simple CNN baseline for comparison
4. Training Setup: Configure optimizers, schedulers, and training loops
5. Model Training: Train both models for 50 epochs
6. Evaluation:
   • Learning curves
   • Confusion matrices
   • Misclassified examples
7. Results Comparison: Detailed analysis of ResNet18 vs Simple CNN
8. Conclusion: Summary and future work suggestions

The notebook provides:

• Training and validation accuracy/loss curves
• Final test accuracy for both models
• Per-class accuracy breakdown
• Confusion matrices showing common misclassifications
• Visual examples of misclassified images
• Comprehensive performance comparison

• 18-layer deep residual network
• Residual connections (skip connections) every 2 layers
• Batch normalization after each convolution
• ~11 million parameters

• 3 convolutional layers with max pooling
• 2 fully connected layers
• Dropout for regularization
• ~0.5 million parameters

• Optimizer: SGD with momentum (0.9)
• Learning Rate: 0.1 with MultiStepLR scheduler
• Weight Decay: 1e-4
• Batch Size: 128
• Epochs: 50
• Loss Function: Cross Entropy Loss

• Random horizontal flip (p=0.5)
• Random crop (32x32 with padding=4)
• Normalization using CIFAR-10 statistics

1. Transfer learning with pre-trained weights
2. Advanced data augmentation (AutoAugment, CutMix)
3. Experiment with ResNet34, ResNet50
4. Apply to CIFAR-100 or ImageNet subsets
5. Model compression and optimization
6. Attention mechanisms (CBAM, SE blocks)

• He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. CVPR.
• Krizhevsky, A. (2009). Learning multiple layers of features from tiny images. Technical Report.
• PyTorch Documentation: https://pytorch.org/docs/stable/index.html

This project is for educational purposes.

Created as part of a machine learning project for CIFAR-10 image classification.