This code trains a convolutional neural network on the CIFAR-10 image dataset. Uses the Keras and NumPy libraries.
INPUT -> [CONV -> RELU -> CONV -> RELU -> POOL]*3 -> [FC -> RELU]*2 -> FC
As per suggestion by Stanford's CS231n
This model has achieved 70.89% test accuracy on the CIFAR-10 dataset with batch size of 128 and 30 epochs.
Maybe not so impressive, but it was fun to build.
sudo pip3 install -r requirements.txt
python3 c10cnn.py