main

We described ToyResNet in Keras and saved model configuration.

It used simple initializers and training from zero is too long.

It's better to load pretrained model.

Just loading ToyResNet trained in Keras.

So, let's update the weights of last layer of the pretrained model from Keras.

All layers except last should be frozen.

As a result the training will be fast.

Just loading ToyResNet trained in Keras, making a copy and using for prediction.