We found that the model that we developed has about 16 layers and that its validation accuracy is about 80%. From this, we observed how the neural network sees the images at different layers. This brings up two questions:
- How does our custom neural network compare with the more advanced neural network?
- How does the advanced neural network see an image compared to our custom neural network? Do all neural networks see an image in a similar way or differently?
To answer these questions, we will train our classifier against two advanced networks, VGG16 and InceptionV3, and visualize a chair image at different layers of the network. Chapter 5, Neural Network Architecture ...
