Image Classification

Implementation and investigation of Fully Connected and Convolutional neural networks on the CIFAR-10 dataset.

• Fully Connected Neural Network
  • Stochastic Gradient Descent
  • SGD with Momentum
  • RMSProp
  • Adam
• Batch Normalization
• Dropout
• Convolutional Neural Network

Setup

Install virtual environment:

sudo pip install virtualenv      # This may already be installed
virtualenv .env                  # Create a virtual environment
source .env/bin/activate         # Activate the virtual environment
pip install -r requirements.txt  # Install dependencies

deactivate                       # Exit the virtual environment

Download data: Run the following to download the CIFAR-10 dataset:

cd deeplearning/datasets
./get_datasets.sh

Compile the Cython extension: Convolutional Neural Networks require a very efficient implementation. You will need to compile the Cython extension before you can run the code. From the deeplearning directory, run the following command:

python setup.py build_ext --inplace