Visualizing convnet filters can't get the results as expected

[ZEROYXY]

This issue occured when I learned the Visualizing convnet filters. I tried the code as below but the grid didn't show the resluts as it said. I get the table with margin but empty in each grid. I tried to debug and thought the issue may happen when the initialization of the 'resullts' or when the 'filter_image' put into the 'resluts'. But I didn't found the root cause. Can anyone give a hand to update it? The pyhton codes are as below:

from keras.applications import VGG16 from keras import backend as K import numpy as np import matplotlib.pyplot as plt

model = VGG16(weights='imagenet', include_top=False)

def deprocess_image(x): x -= x.mean() x /= (x.std()+1e-5) x = 0.1 x += 0.5 x = np.clip(x, 0, 1) x = 255 x = np.clip(x, 0, 255).astype('uint8') return x

def generate_pattern(layer_name, filter_index, size=150): layer_output = model.get_layer(layer_name).output loss = K.mean(layer_output[:, :, :, filter_index]) grads = K.gradients(loss, model.input)[0] grads /= (K.sqrt(K.mean(K.square(grads)))+1e-5) interate = K.function([model.input], [loss, grads]) input_img_data = np.random.random((1, size, size, 3))20+128. step = 1. for i in range(40): loss_value, grads_value = interate([input_img_data]) input_img_data += grads_valuestep img = input_img_data[0] return deprocess_image(img)

plt.imshow(generate_pattern('block3_conv1', 0)) plt.show()

for layer_name in ['block1_conv1', 'block2_conv1', 'block3_conv1', 'block4_conv1']: size = 64 margin = 5 results = np.zeros((8size+7margin, 8size+7margin, 3)) for i in range(8): for j in range(8): filter_image = generate_pattern(layer_name, i+(j8), size=size) horizontal_start = isize+imargin horizontal_end = horizontal_start+size vertical_start = jsize+j*margin vertical_end = vertical_start+size results[horizontal_start:horizontal_end, vertical_start:vertical_end, :] = filter_image plt.figure(figsize=(20, 20)) plt.imshow(results) plt.show()

The screenshot of the issue is as below:

[OndrejSliva]

Hi, allowed range of values for imshow parameter is [0..1] for floats or [0..255] for integers. Data type of variable result is float, but range of values is [0..255]. So you need to change data type to int, or divide by 255.

First solution: plt.imshow(results.astype(np.uint8))

Second solution: plt.imshow(results/255)

[ZEROYXY]

Hi Silva

^_^ Thank you very much for your time to answer my question and I checked the parameter and code as you suggested.  Then I found the key point you said and I tried your way to make this issue resolved. Thank you so much again and I can close the CNN Fillters Visualization successfully now with your strong help ! I will tell the people who is facing the same problem with me about how to resolve it, too. 

Thanks and Best Regards Cloud [Shuo Zhang] Your Chinese Friend

------------------ 原始邮件 ------------------ 发件人: "OndrejSliva"<notifications@github.com>; 发送时间: 2020年4月5日(星期天) 晚上8:14 收件人: "fchollet/deep-learning-with-python-notebooks"<deep-learning-with-python-notebooks@noreply.github.com>; 抄送: "西伯利亚狼"<845605918@qq.com>;"Author"<author@noreply.github.com>; 主题: Re: [fchollet/deep-learning-with-python-notebooks] Visualizing convnet filters can't get the results as expected (#133)

Hi, allowed range of values for imshow parameter is [0..1] for floats or [0..255] for integers. Data type of variable result is float, but range of values is [0..255]. So you need to change data type to int, or divide by 255.

First solution: plt.imshow(results.astype(np.uint8))

Second solution: plt.imshow(results/255)

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[zengxinch]

Yes, you could add the chapter number to the question title. I think this could help people who encounter the same problem find answer quickly.

[SaadAsgharAli]

Thanks @OndrejSliva