一文带你用可视化理解卷积神经网络(18)


image.shape[1
image.shape[2
)))[0
[correct_class

print(xyn' - 'image.shape)
heatmap1 = heatmap/heatmap.max()
plt.imshow(heatmap)

这非常有意思 。 我们现在将使用标准化的热图概率创建一个掩模并绘制它:

import http://skimage.io as io
#creating mask from the standardised heatmap probabilities
mask = heatmap1 < 0.85
mask1 = mask *256
mask = mask.astype(int)
io.imshow(maskcmap='gray')

最后 , 我们将遮挡码强加在输入图像上并绘制:

import cv2
#read the image
image = cv2.imread('car.jpeg')
image = cv2.cvtColor(imagecv2.COLOR_BGR2RGB)
#resize image to appropriate dimensions

推荐阅读