Unofficial implementation for Grad-CAM in Pytorch with Multi Network Structures

What makes the network think the image label is 'dog' and 'cat':

Combining Grad-CAM with Guided Backpropagation for the 'dog' class:

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

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

---

In this Repo

Grad-CAM, Guided-Backpropagation with Grad-CAM

VGG19

---

VGG 19 Layer1, Layer20 and Layer36

---

EfficientNet-b0

---

EfficientNet-b0 Layer1, Layer10 and Layer15

---

---

What exactly is the difference among this repository and the others?

A: For example, these two are the most popular efficientdet-pytorch,

• Add EfficientNet for Visualization (Can use for both torchvision.models and EfficientNet from lukemelas/EfficientNet-PyTorch)
• Add multi-layer visualization for comparison
• Switch GuidedBackPropagationReLU to GuidedBackPropagationSwish for EfficientNet

---

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

Reference: Appreciate the great work from the following repositories:

• lukemelas/EfficientNet-PyTorch
• jacobgil/pytorch-grad-cam