TJJTJJTJJ / pytorch-grad-cam

PyTorch implementation of Grad-CAM

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Grad-CAM implementation in Pytorch

What makes the network think the image label is 'pug, pug-dog' and 'tabby, tabby cat':

Dog Cat

Gradient class activation maps are a visualization technique for deep learning networks.

See the paper: https://arxiv.org/pdf/1610.02391v1.pdf

The paper authors torch implementation: https://github.com/ramprs/grad-cam

My Keras implementation: https://github.com/jacobgil/keras-grad-cam

The interpretation:

https://www.jianshu.com/p/b2f7efe10ad8

https://tjjtjjtjj.github.io/2018/12/14/Grad-CAM/


This uses VGG19 from torchvision. It will be downloaded when used for the first time.

The code can be modified to work with any model. However the VGG models in torchvision have features/classifier methods for the convolutional part of the network, and the fully connected part. This code assumes that the model passed supports these two methods.


Usage: python grad-cam.py --image-path <path_to_image>

To use with CUDA: python grad-cam.py --image-path <path_to_image> --use-cuda

About

PyTorch implementation of Grad-CAM


Languages

Language:Python 100.0%