visualize filter by gradient descent on a random image

visulize_filter_by_optimization.py

from keras import backend as K

img_width, img_height = 28,28 

def visualize_filter(model, layer_name = 'convolution2d_1', filter_index = 0):
    first_layer = model.layers[0]
    # this is a placeholder tensor that will contain our generated images
    input_img = first_layer.input

    # get the symbolic outputs of each "key" layer (we gave them unique names).
    layer_dict = dict([(layer.name, layer) for layer in model.layers])
    # build a loss function that maximizes the activation
    # of the nth filter of the layer considered
    layer_output = layer_dict[layer_name].output
    loss = K.mean(layer_output[:, filter_index, :, :])

    # compute the gradient of the input picture wrt this loss
    grads = K.gradients(loss, input_img)[0]

    # normalization trick: we normalize the gradient
    grads /= (K.sqrt(K.mean(K.square(grads))) + 1e-5)

    # this function returns the loss and grads given the input picture
    iterate = K.function([input_img], [loss, grads])
    import numpy as np

    
    # we start from a gray image with some noise
    input_img_data = np.random.random((1, 1, img_width, img_height)) * 20 + 128.
    # run gradient ascent for 20 steps
    step = 1.0
    for i in range(20):
        loss_value, grads_value = iterate([input_img_data])
        input_img_data += grads_value * step

    return input_img_data.reshape((img_width, img_height))

# usage
nb_filter = 10
for i in range(nb_filter):
    img = visualize_filter(model, layer_name = 'convolution2d_2', filter_index = i)
    plt.figure()
    plt.imshow(img)