Class Activation Mapping visulaization

Class Activation Mapping is a way to enables the convolutional neural network to have remarkable localization ability despite being trained on image-level labels. I use it to visualize what my model is looking in the images.

The theory is described here: http://cnnlocalization.csail.mit.edu/

The script cam_keras.py is the Keras implementation of Class Activation Mapping. It is based on this script in pytorch.

Installing Dependencies

This project depends on latest version of

• tensorflow
• pillow
• opencv-python
• numpy
• requests

Do pip install <missing_package> to install whatever you don’t have. Using conda environment is recommended.

Running cam_keras

To get the output you have to run:

python3 cam_keras.py

Note:

1. To change the model, change the model_id in the script.
2. The output image is stored in CAM.jpg.
3. To change the input image, modify the IMG_URL in the script.

Example result

Input image

Output image