Understanding tensorboard images

Question

I want to use keras+tensorboard. My architecture looks like this:

tbCallBack = TensorBoard(log_dir='./logs', histogram_freq=2, batch_size=32, write_graph=True, write_grads=True, write_images=True)

K.clear_session()
sess = tf.Session()
K.set_session(sess)

input_img = Input(shape=(augmented_train_data[0].shape[0], augmented_train_data[0].shape[1], 3))

x = Conv2D(8, (1, 1), padding='same', activation='relu', name="1x1_1")(input_img)
x = Conv2D(16, (3, 3), padding='same', activation='relu', name="3x3_1")(x)
x = Conv2D(32, (3, 3), padding='same', activation='relu', name="3x3_2")(x)
x = Conv2D(1, (1, 1), padding='same', activation='relu', name="1x1_2")(x)
x = Flatten()(x)
x = Dense(16, activation='relu')(x)


output = Dense(2)(x)

model = Model(inputs=input_img, outputs=output)
model.compile(optimizer='adam', loss='mean_squared_error')

#tbCallBack.set_model(model)
print(model.summary())

history = model.fit(augmented_train_data, augmented_train_label, validation_data=[augmented_validation_data, augmented_validation_label] ,epochs=20, batch_size=32, callbacks=[tbCallBack])

When looking at the tensorboard image tab, it looks like this I cant quite interpret that though, I thought this tab would show how the weights of my convolutions develop over the epochs. So, how to interpret these images. Or did I do a mistake in setting up tensorboard?

Answer 1

It looks like that is exactly what you are getting. The grayscale of the image shows the weights. The slider on top can be used to go back and forth in epochs and hence look at the training progression.