Convolutional Neural Networks (CNN) for Image Classification

This repository contains implementations of Convolutional Neural Networks (CNNs) applied to the CIFAR-10 dataset for image classification.

Project Overview

Convolutional Neural Networks (CNNs) are powerful deep learning models designed to learn spatial hierarchies of features from images.
They are widely used in computer vision tasks such as object detection, facial recognition, and medical image analysis.

This project explores:

• A CNN implemented with TensorFlow/Keras
• Performance improvements using Batch Normalization, Dropout, and Fully Connected Layer tuning

---

Dataset

• CIFAR-10 dataset
• 60,000 color images (32x32 pixels, 10 classes)
  • 50,000 training images
  • 10,000 test images
• Classes: Airplane, Automobile, Bird, Cat, Deer, Dog, Frog, Horse, Ship, Truck

---

CNN Architecture

Baseline CNN (TensorFlow/Keras)

• Conv Layer 1: 32 filters (3×3), ReLU
• Conv Layer 2: 32 filters (3×3), ReLU
• Max Pooling: 2×2
• Conv Layer 3: 64 filters (3×3), ReLU
• Conv Layer 4: 64 filters (3×3), ReLU
• Max Pooling: 2×2
• Conv Layer 5: 128 filters (3×3), ReLU
• Max Pooling: 2×2
• Flatten
• Dense Layer: 256 neurons, ReLU
• Output Layer: Softmax (10 classes)
• Optimizer: Adam (learning rate = 0.001)
• Loss: Sparse categorical cross-entropy
• Training: 20 epochs, batch size = 128

---

Results

Model	Accuracy	Loss
Baseline CNN	70.64%	1.3005
Improved CNN	79.33%	0.6270

---

Improvements

1. Batch Normalization

   • Normalizes activations after convolution layers.
   • Stabilizes training and speeds convergence.

2. Dropout

   • Prevents overfitting by randomly dropping neurons during training.
   • Applied: 25% after convolutional layers, 50% after dense layer.

3. Increased Neurons in Dense Layer

   • Increased from 256 → 512 for improved learning capacity.

4. Learning Rate Tuning

   • Adam optimizer with learning rate = 0.001 improved convergence.

---

Transfer Learning (Bonus)

• Implemented ResNet18 using PyTorch
• Modified final fully connected layer (512 → 10 for CIFAR-10 classes)
• Optimizer: SGD, learning rate = 0.001

---

Conclusion

• CNN on CIFAR-10 achieved 70.64% accuracy baseline.
• With improvements (BatchNorm, Dropout, FC tuning, LR tuning), accuracy increased to 79.33%.

---

Requirements

Install dependencies:

pip install numpy tensorflow keras torch torchvision matplotlib

---

How to Run

1. Clone the repository:

   git clone https://github.com/harsh7z/cnn-image-classification.git
   cd cnn-image-classification

2. Run the notebooks:

   jupyter notebook hw4.ipynb

3. Train models and evaluate performance.