This is the codes for reproducing the results in my paper, Visualizing Deep Neural Network by Alternately Image Blurring and Deblurring.
http://www.sciencedirect.com/science/article/pii/S0893608017302095
The codes are not well organized and need some specail skills to tune the hyper-paramters, such as learning rate and the gaussian std of the blur kernel. I am attempting to reduce the parameters or find a way to automatically set them.
AlexNet:
VGG:
GoogLeNet conv2:
GoogLeNet Inceptions:
