Image Classification using CNN on CIFAR-10

Overview

This project implements a Convolutional Neural Network (CNN) for image classification using the CIFAR-10 dataset. The model is trained to classify images into 10 categories and achieves an accuracy of 77.21%.

Dataset

CIFAR-10 contains 60,000 color images (32×32 pixels) belonging to the following classes:

• Airplane
• Automobile
• Bird
• Cat
• Deer
• Dog
• Frog
• Horse
• Ship
• Truck

Dataset Split:

• Training Images: 50,000
• Testing Images: 10,000

Technologies Used

• Python
• PyTorch
• Torchvision
• NumPy
• Matplotlib
• Scikit-learn

CNN Architecture

• Conv2D (32 Filters)
• ReLU
• Max Pooling
• Conv2D (64 Filters)
• ReLU
• Max Pooling
• Fully Connected Layer (512 Neurons)
• Dropout
• Output Layer (10 Classes)

Training Configuration

• Optimizer: Adam
• Learning Rate: 0.001
• Epochs: 20
• Batch Size: 64
• Loss Function: CrossEntropyLoss

Performance

• Accuracy: 77.21%
• Precision: 77%
• Recall: 77%
• F1-Score: 77%

Results

Generated Outputs:

• Training Loss Curve
• Confusion Matrix
• Cat and Dog Sample Images
• Trained CNN Model

Project Structure

ComputerVision_CNN_Project/

• model.py
• train.py
• evaluate.py
• show_samples.py
• requirements.txt
• README.md

results/

• loss_curve.png
• confusion_matrix.png
• cat_dog_samples.png

saved_models/

• cnn_cifar10.pth

How to Run

Install dependencies:

pip install -r requirements.txt

Train the model:

python train.py

Evaluate the model:

python evaluate.py

Display sample images:

python show_samples.py

Author

Muhammad Aman Fasihe

BS Artificial Intelligence

The Islamia University of Bahawalpur

Note: The trained model file (.pth) is included in the project submission package. The GitHub repository contains the source code, documentation, and result visualizations.