I would like to visualise the first Convolutional Layer and its weights, and to do so

Bit of a better approach here. <div class="highlight highlight-source-python notra

Visualise first Convolutional Layer about keras-extra HOT 2 CLOSED

anayebi commented on July 22, 2024

Visualise first Convolutional Layer

from keras-extra.

Comments (2)

KeironO commented on July 22, 2024

To any researchers reading this, sleep is important. Go take a nap.

So, turns out that this was a very silly question.

Here's a horrible hacky way of doing it for now, I'll work on something better - soonish.

def visualise_first_layer(model, convout1, test_data):
    get_layer_output = K.function([model.layers[0].input], [convout1.get_output(train=False)])
    layer_output = get_layer_output([test_data])[0]

    f, (ax1, ax2, ax3, ax4, ax5, ax6) = plt.subplots(1, 6, sharey=True)
    ax1.imshow(layer_output[0,0,1,:])
    ax2.imshow(layer_output[0,10,1,:])
    ax3.imshow(layer_output[0,20,1,:])
    ax4.imshow(layer_output[0,30,1,:])
    ax5.imshow(layer_output[0,40,1,:])
    ax6.imshow(layer_output[0,50,1,:])

    plt.show()

from keras-extra.

KeironO commented on July 22, 2024

Bit of a better approach here.

def visualise_first_layer(model, convout1, test_data):
    get_layer_output = K.function([model.layers[0].input], [convout1.get_output(train=False)])
    layer_output = get_layer_output([test_data])[0]

    for i in range(layer_output.shape[1]):
        p = plt.imshow(layer_output[0,i,1,:])
        plt.pause(0.1)