Deep learning starts from scratch-neural network (seven and one-half) convolution optimization structure, cat and dog recognition, optimization step results

Directly four groups of convolutional layers and maximum pool words, one layer classification and one layer result

After adding data enhancement:

Fast feature extraction without data augmentation

Add your own classifier after VG16 (two Dense, one dropout regularization)

VGG16 freezes and uses data-enhanced feature extraction.

VGG16 freezes, uses data-enhanced feature extraction, and after training the classifier, fine-tune the layer above block5_conv1 of the model.