Evolving-AI-Lab / fooling

Code base for "Deep Neural Networks are Easily Fooled" CVPR 2015 paper

171 stars 59 forks source link

no attribute 'backward_from_layer' #2

Closed xiangyangli-cn closed 9 years ago

xiangyangli-cn commented 9 years ago

when i run find_fooling_image.py This came: AttributeError: 'Classifier' object has no attribute 'backward_from_layer'. How can i fix it?

anguyen8 commented 9 years ago

@lxycnu : If you run the gradient ascent code, please install the lastest Caffe which supports that attribute backward_from_layer. The Caffe version included in this repo is only for the evolutionary code.

xiangyangli-cn commented 9 years ago

I have tried the current version, but it also does not support the attribute of ' backward_from_layer. Maybe i can use 'backward' instead of '(Net.backward = _Net_backward) backward_from_layer '.

xiangyangli-cn commented 9 years ago

backout = net.backward_from_layer(push_layer, diffs)

backout=net.backward(diffs,push_layer)

...... I change it like this, but it still has some problem. Can you give me some suggestion?

anguyen8 commented 9 years ago

@yosinski : Can you share which Caffe version you used to have working with find_fooling_image.py?

yosinski commented 9 years ago

@anguyen8 @lxycnu Oops, I see that I forgot to push that modification to Caffe. I'll push it ASAP.

yosinski commented 9 years ago

@lxycnu Try again using these updated instructions and the ascent branch. I just tried and it works for me.