This project is aiming to learn more about neural networks and especially convolutional neural networks. It aims to recognize images such as humans, animals, vehicules and objects. It is fully implemented in C and aims to be fast and memory efficient.

Each data passed to the neural network has to be normalized to maintain consistency over the learning of such a neural network. The choosen size is 300x300 pixels.

At the moment, the forward propagation is working just fine.I am still trying to implement a correct back propagation though.

https://github.com/libsdl-org/SDL/releases/tag/release-2.24.1

./install-dep.sh

python3 resize-image.py ${IMAGE_TO_NORMALIZE}

Not implemented yet.

To try the network on one of your image, just run

make

./CNN [--verbose] ${PATH-TO-IMAGE}

Verbose option allows to see every step of the prediction process including convolution and pooling outputs. Don't forget to normalize your image.