为什么pycaffe向后的结果总是0？

Question

net.blobs['data'].data[...] = transformed_image
output = net.forward()

output_prob = output['prob'][0]  # the output probability vector for the 
first image in the batch

print 'predicted class is:', output_prob.argmax()

label_index = output_prob.argmax()
caffeLabel = np.zeros((1,1000))
caffeLabel[0,label_index] = 1;

vis_layer = 'pool5' # visualization layer

grads=net.backward(diffs=[vis_layer],**{'prob':caffeLabel})
print(np.sum(grads))

我想用这种方式获取渐变，但是print(np.sum(grads))总是0，我换了图层conv5或者其他图层，都没用！

Answer 1

我已经解决了这个问题，将以下代码添加到 'deploy.prototxt'

force_backward:true