Caffe/+Interfaces fesibility towards intermediate gradient computations

[Rajasekhar Ponakala]

Hi

I'm trying to implement the framework described in DeepXplore (https://github.com/peikexin9/deepxplore)[Python-based]. DeepXplore uses Tensorflow+Keras. Following are the feasibility doubts I've.

[1]. Can we compute gradient of any intermediate neuron w.r.t. input tensor using Caffe [joint optimization]?
[2]. How feasible the Caffe on implementing VGG-16, VGG-19, ResNet-50 w.r.t [1]?
[3]. If [1] yes, can we implement with ImageNet Caffe model?

Thank you.

[Rajasekhar Ponakala]

adding [4] as with GoogleNet?

[huwei...@gmail.com]

hello have you save the problem in caffe? I am doing this work recently, any help will be appreciated

在 2018年6月6日星期三 UTC+8上午1:44:00，Rajasekhar Ponakala写道：