Holistic-CAM: Ultra-lucid and Sanity Preserving Visual Interpretation in Holistic Stage of CNNs

Pytorch implementation of Holistic-CAM: Ultra-lucid and Sanity Preserving Visual Interpretation in Holistic Stage of CNNs, which is accepted by ACM MM2024.

😄 Pipeline of Holistic-CAM:

🔥 Compare with SOTA methods:

🛠️ Requirements:

python 3.X
jupyter notebook
pytorch >= 1.5 (including torchvision)
matplotlib
opencv-python

🗝️ How to Run:

use demo.ipynb

✏️ Citation:

@inproceedings{chen2024holisticcam, 
   title={Holistic-{CAM}: Ultra-lucid and Sanity Preserving Visual Interpretation in Holistic Stage of {CNN}s}, 
   author={Pengxu Chen and Huazhong Liu and Jihong Ding and Jiawen Luo and Peng Tan and Laurence T. Yang}, 
   booktitle={Proceeding of the ACM Intertional Conference on Multimedia}, 
   year={2024},
   pages={5423–5431},
   doi={10.1145/3664647.3681707}
}

📁 Supplementary material:

Due to the page limitations of MM, we have placed the remaining ablation stduy and complete experimental results in the supplementary.pdf.

✉ Contact:

Any questions can be asked at [email protected]