A CNN created from scratch using numpy only

I created this model while working on my thesis to help me build deep understanding of CNN.

This is also trained using degree 2 polynomial approximation of ReLU a part from simpler ReLU for gathering weights and biases for my Encrypted CNN. There are versions present of stride 2 and 3 for the purpose of better implementation on the fully homomorphic encrypted data.

This is a very slow version as I used loops to perform convolution and pooling operation. But it can provide good understanding.