Visualize the layer activations of a Convolutional Neural Network to better understand the models behavior.

via_activations.py

import numpy as np
from keras.models import Model

conv_layers = [layer for layer in my_model.layers if 'Conv' in type(layer).__name__]
layer_outputs = [layer.output for layer in conv_layers]

activation_model = Model(inputs=my_model.input, outputs=layer_outputs)
activations = activation_model.predict(np.expand_dims(img, axis=0))

ncols = 5
layer_activationss = [[np.squeeze(layer) for layer in np.split(
  np.squeeze(activation), activation.shape[-1], axis=2)] for activation in activations]

for layer, (layer_activations, layer) in enumerate(zip(layer_activationss, conv_layers)):
  nrows = int(np.ceil(len(layer_activations) / ncols))
  fig, ax = plt.subplots(nrows, ncols, figsize=(20, nrows * 4))
  for i in range(nrows):
    for j in range(ncols):
      if i * j + j < len(layer_activations):
        ax[i, j].imshow(layer_activations[i * j + j], cmap='jet')

      ax[i, j].axes.get_xaxis().set_visible(False)
      ax[i, j].axes.get_yaxis().set_visible(False)
      
  print('Visualizing layer {}: {} with {} feature maps'.format(
    layer, type(layer).__name__, len(layer_activations)))
  plt.show()