To appear at IEEE CVPR 2020 Workshop on Fair, Data Efficient and Trusted Computer Vision.
In this paper, we develop a novel post-hoc visual explanation method called Score-CAM based on class activation mapping. Score-CAM is a gradient-free visualization method, extended from Grad-CAM and Grad-CAM++. It achieves better visual performance and fairness for interpreting the decision making process.
Paper: Score-CAM: Score-Weighted Visual Explanations for Convolutional Neural Networks (Haofan Wang, Zifan Wang, Mengnan Du, Fan Yang, Zijian Zhang, Sirui Ding, Piotr Mardziel and Xia Hu.)
2020.4.13
: First version of Score-CAM code has been released. More implementations will be added later.
-
- Support for Colab notebook.
-
- Support for faster version of Score-CAM.
-
- Support for pre-trained model in Pytorch.
-
- Support for self-defined model in Pytorch.
-
- Add visualization result and quantitive evaluation.
-
- Support for other tasks such as object localization task.
It would be very appreciated for implementing Score-CAM for other popular projects, if any of you are interested.
-
- isses #350, Implement in pytorch/captum
-
- issues #124, Implement in sicara/tf-explain
-
- Implement in PAIR-code/saliency
-
- Implement in experiencor/deep-viz-keras
-
- Implement in utkuozbulak/pytorch-cnn-visualizations
Before we release the official code, some great researchers have implemented Score-CAM on different framework. I am very grateful for the efforts made in their implementation.
Demystifying Convolutional Neural Networks using ScoreCam
kerasでScore-CAM実装.Grad-CAMとの比較
If you find this work or code is helpful in your research, please cite and star:
@article{wang2019scorecam,
title={Score-CAM: Score-Weighted Visual Explanations for Convolutional Neural Networks},
author={Wang, Haofan and Wang, Zifan and Du, Mengnan and Yang, Fan and Zhang, Zijian and Ding, Sirui and Mardziel, Piotr and Hu, Xia},
journal={arXiv preprint arXiv:1910.01279v2},
year={2019}
}
Utils are built on flashtorch, thanks for releasing this great work!
If you have any questions, feel free to contact me via: haofanw@andrew.cmu.edu