Grad-CAM for GoogLenet and ResNet

Implementation of grad-cam for GoogLenet and ResNet architechtures

Setup

• Fullfill dependencies as required by the author's repository
• Clone the repository
  • git clone --recursive https://github.com/utsavgarg/gradcam-googlenet-resnet.git
• For ResNet-101, download model from here and place it in fb.resnet.torch directory
• Change two path to the following, in inception.torch/googlenet.lua
  • local w = paths.concat('inception.torch/' .. f, m.gname .. '_w.h5')
  • local b = paths.concat('inception.torch/' .. f, m.gname .. '_b.h5')

Run

• For GoogLenet
  • th gcam_googleNet.lua
• For ResNet 101
  • th gcam_resNet.lua

Sample

Input	GoogLeNet	ResNet

About

• This repository is a minor modification over the code provided by authors here to add support for GoogLenet and ResNet-101 architectures, and most files are modifications over the original code provided.
• GoogLenet architecture and weights taken from here
• ResNet-101 model and image preprocessing are from here

Licence

BSD