CIFAR-10 Image Classification with Convolutional Neural Networks (CNN)

Description

This project explores, develops, and evaluates various CNN architectures trained on CIFAR-10 dataset.

The objective is to build CNN models in aim to experiment and explore various deep learning neural network architectures, model optimisation and regularisation techniques, hyperparameter tuning, and data augmentation to accurately predict the class of each image while ensuring models do not exhibit overfitting.

In this initial approach, we achieved a prediction accuracy of 75-79%, and these models accuracies can be further improved with additional training.

Models

• Basic CNN model with 2 convolutional blocks
• Basic CNN model with 3 convolutional blocks
• Optimised deeper model with 3 convolutional blocks
• Residual Network (ResNet34) model
• Very Deep Convolutional Network (VGG) model

Prerequisites

• Python 3.x
• Pip: Python package installer, alternatively you could use Conda.
• Libraries:
  • PyTorch
  • Torchvision
  • NumPy
  • Matplotlib

License

This project is licensed under the MIT License.