activation_maximization.py

import numpy as np
import keras
from keras.datasets import mnist
from keras.models import Sequential, Model,load_model
from keras.layers import Dense, Dropout, Flatten, Activation, Input
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K

## Load the model the see the activation map
model = load_model('handwritten.h5')
print('hello')

# visualize the dense layer
from vis.visualization import visualize_activation
from vis.utils import utils
from keras import activations
from matplotlib import pyplot as plt

plt.rcParams['figure.figsize'] = (18, 6)

# Utility to search for layer index by name. 
# Alternatively we can specify this as -1 since it corresponds to the last layer.
layer_idx = utils.find_layer_idx(model, 'preds')

# Swap softmax with linear
model.layers[layer_idx].activation = activations.linear
model = utils.apply_modifications(model)

# This is the output node we want to maximize.
filter_idx = 0
img = visualize_activation(model, layer_idx, filter_indices=filter_idx)
plt.imshow(img[..., 0])
plt.savefig('foo.png')

import string

s = string.ascii_uppercase

for output_idx in np.arange(26):
    img = visualize_activation(model, layer_idx, filter_indices=output_idx, input_range=(0., 1.))
    plt.figure()
    plt.title(f'Networks perception of {s[output_idx]}')
    plt.imshow(img[..., 0])
    plt.savefig(f'{output_idx}.png')