GradCAM-pytorch

To activate map for Gradient descent weights in deep learning models

Available Models

• ResNet-101
• Inception-ResNet-V2
• EfficientNet
• NASNet-A-Large
• DenseNet-161
• DenseNet-169
• ...

Code Explanation

Main Code List:

• gradcam.py
• gradcam_for_video.py
• models.py
• CLAHE_Aug.py

Folder and Code contents

Codes

• gradcam.py

  • Image folderlist use making GradCAM images
  • folder name has label
  • GradCAM image name is groundtruth_prediction_probability_imagename.jpg

• gradcam_for_video.py

  • video image frame use making GradCAM images

• models.py

  • selected model definition code

• CLAHE_Aug.py

  • video image frame convert CLAHE image

Folders

• CAM

  • it saves CAM output image

• Model

  • Saving models to predict probability and output image

• test

  • test set images to predict

  ## folder structure
    +-- test
    |   +-- class1
    |           +-- img1.jpg
    |           +-- img2.jpg
    |           +-- img3.jpg
    |   +-- class2
    |   +-- class3
  

Contents

• This GradCAM codes read PIL image library
• All I want various models can make GradCAM

Examples

How to run

all gradcam code should control main function

• python gradcam.py
• python gradcam_for_video.py

Reference

• main refernce
  • https://github.com/kazuto1011
• I modified hyunseok oh codes
  • https://github.com/hyunseokOh