3 different types of convolutional neural networks to classify cats and dogs

This model is created from nothing. It contains 4 Convolutional layers, each of them followed by a MaxPooling. The model then flattens the result and passes it through a dense layer to final give to the final sigmoid layer. This model uses Data Augmentation and a dropout (right after flattening the last convolutional layer) to avoid overfitting.

This model uses the pretrained VGG16 for convolutional, but add its own dense layers (same of the previous one). It does not uses Data Augmentation. None of the weights of the VGG16 model are modified.

This model is based on the previous one, but allows for Data Augmentation. The weights of the last layers of VGG16 are modified (block 5, containing 3 convolutional layers and 1 MaxPooling).