heatmap-視覺化你的模型辨識特徵

本模型使用ResNet20訓練cifar10資料集保存的權重。

載點點我
論文出處:Grad-CAM Visual Explanations from Deep Networks via Gradient-Based Localization
簡單來說cnn可以萃取圖片的特徵，越深的捲積層，越可以表現重要，可讓神經網路學習的特徵，Grad-CAM的目的就是提取最後一層的特徵，並且resize為原圖的大小，這就可以視覺化你的模型辨識特徵


程式碼如下:

import cv2
import numpy as np  
from keras.models import load_model  #用來啟動 NN
from keras import backend as K
from  matplotlib import pyplot as plt
test_image = cv2.resize(cv2.imread('./plane1.jpg'),(32,32))
test_image = np.asarray(test_image.astype("float32"))
test_image = test_image/255.
test_image = test_image.reshape((1,32,32,3))
model=load_model('./resnet_softmax.h5')
preds = model.predict(test_image)
print("所有結果resnet_softmax=",preds)
print("第"+str(np.argmax(preds[0])+1)+"類") #返回數值最大的索引
#model.summary()
model_output = model.output[:, np.argmax(preds[0])]
last_conv_layer = model.get_layer('conv2d_22') 
grads = K.gradients(model_output, last_conv_layer.output)[0]
pooled_grads = K.mean(grads, axis=(0, 1, 2))
iterate = K.function([model.input], [pooled_grads, last_conv_layer.output[0]])
pooled_grads_value, conv_layer_output_value = iterate([test_image])
for i in range(256): #256是conv2d_22的維度
    conv_layer_output_value[:, :, i] *= pooled_grads_value[i]

heatmap = np.mean(conv_layer_output_value, axis=-1)
#這裡你可以print一下heatmap
#print(heatmap)
heatmap = np.maximum(heatmap, 0)
heatmap /= np.max(heatmap)
'''plt.matshow(heatmap)
plt.savefig('bob1.png')
plt.show()'''
img = cv2.imread('./plane1.jpg')
heatmap = cv2.resize(heatmap, (img.shape[1], img.shape[0]))
heatmap = np.uint8(255 * heatmap)
heatmap = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET)
superimposed_img = heatmap * 0.4 + img
cv2.imwrite('23.jpg', superimposed_img)
imgheat=cv2.imread('23.jpg')
cv2.imshow('imgheat', imgheat)
cv2.waitKey(0)
cv2.destroyAllWindows()