Project for Computational intelligence and Deep Learning, AIDE course at University of Pisa. We constructed multiple models from the ground up, varying their complexity. We utilized various techniques, such as depthwise separable convolution, residual modules, and inception modules. Additionally, we tested three different pretrained models: MobileNetV2, ResNet50V2, and VGG16.