CNN Computer Vision Projects

A collection of deep learning projects focused on image classification using Convolutional Neural Networks (CNNs) built with PyTorch.

Projects

1. Australian Road Sign Classification

A CNN model achieving 93.48% accuracy in classifying Australian road signs into 5 categories:

• Keep Left
• No Left Turn
• No Right Turn
• Pedestrian Crossing
• Speed Limit

Key Features:

• Custom CNN architecture with progressive feature extraction (3→32→64→128 channels)
• Dataset of 4,399 images from Melbourne driving footage
• Strong per-class performance (F1-scores ranging from 88.89% to 96.15%)

2. CIFAR-10 Classification

A CNN implementation achieving 82.61% accuracy on the CIFAR-10 dataset, featuring:

• Comprehensive data augmentation pipeline
• Batch normalization and dropout regularization
• Progressive feature extraction architecture

Requirements

• Python 3.x
• PyTorch >= 2.0.0
• torchvision
• matplotlib
• numpy
• seaborn
• scikit-learn

Getting Started

Each project contains detailed documentation in its respective directory:

• Road Sign Classification
• CIFAR-10 Classification

Results

Project	Accuracy	F1-Score
Road Signs	93.48%	92.98%
CIFAR-10	82.61%	-