Request for the code of evaluation metrics (e.g., relative confidence drop, localization score)

[1215603580-cmyk]

Hi authors,
I’m deeply interested in your work Finer-CAM: Spotting the Difference Reveals Finer Details for Visual Explanation published in CVPR 2025. The method’s design for fine-grained visual explanation is really insightful!
I noticed that the paper provides detailed visualization results and quantitative metrics (like relative confidence drop (RD), deletion AUC, and energy-based pointing game for localization), but the released code repo doesn’t include the implementation of these evaluation metrics yet.
I’m currently working on reproducing and comparing fine-grained XAI methods, and the standard, official implementation of these metrics would be crucial for ensuring the fairness and accuracy of my experiments. Could you please consider open-sourcing the code for these evaluation metrics?
Thank you so much for your contributions to the community and your kind assistance!
Best regards。

[Link7808]

Hi, thanks for reaching out.

For the evaluation metrics you mentioned (e.g., relative confidence drop / RD, deletion AUC, and localization-style scores), you can reference the Quantus library, which provides a well-maintained collection of XAI evaluation metrics:
• Quantus

In my experience, these metrics are also fairly straightforward to implement directly.
If you run into any ambiguity about the exact protocol details, feel free to follow up here with what setup you are using.