Original Implementation of the paper Revisiting The Evaluation of Class Activation Mapping for Explainability: A Novel Metric and Experimental Analysis, appeared at CVPRW'2021 Responsible Computer Vision as oral and poster presentation.
ADCC is a more solid and unbiased benchmark for evaluating CAMs for explainability purposes. It considers more than one aspect of how is the resulting saliency map (read the paper for a more detailed overview)
Given:
- An input image
- A saliency map
- An explanation map
- A CNN
- A saliency map extractor (a callable returning an upsampled saliency map)
it computes the Average Drop, the Coherency and the Complexity, to return the final ADCC score.
Using default parameters
main.py
Using custom parameters
main.py --image [path-to-input-image,str] --model [name-of-the-CNN,str]
The ADCC module provides all the computation needed to return the final score, given the 5 inputs previously mentioned.
The ADCC module simply returns the ADCC score in [0,1] range
This repo is implemented in PyTorch
@inproceedings{poppi2021revisiting,
title={Revisiting The Evaluation of Class Activation Mapping for Explainability: A Novel Metric and Experimental Analysis},
author={Poppi, Samuele and Cornia, Marcella and Baraldi, Lorenzo and Cucchiara, Rita},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
year={2021}
}
For any info contact samuele.poppi@unimore.it