Skip to content

Official Implementation of [ACM-MM'24] Effective Optimization of Root Selection Towards Improved Explanation of Deep Classifiers

License

Notifications You must be signed in to change notification settings

meetxinzhang/XAI-RootInference

Repository files navigation

PyTorch Implementation of Effective Optimization of Root Selection Towards Improved Explanation of Deep Classifiers

fig 1 turbo

Comparative illustration of heatmaps generated by our proposed and the best baseline of root scheme (DTD-Z+) over ImageNet. As seen, the advantages achieved by our proposed can be highlighted as: (i) more accurate explanations with less noise; (ii) stronger hierarchical representation, where different regions of the target object have different colors, reflecting that the weights assigned by the proposed are closer to optimal; (iii) reflection of object contours and even textures. The visualizations utilize the Turbo colormap to highlight details and ensure color (\emph{i.e.} relevance) accessibility for colorblind readers, and mean cropping is applied to enhance the contrast.

Reference Format:

Xin Zhang, Shenghua Zhong, and Jianmin Jiang. 2024. Effective Optimiza- tion of Root Selection Towards Improved Explanation of Deep Classifiers. In Proceedings of the 32nd ACM International Conference on Multimedia (MM’24), October 28-November 1, 2024, Melbourne, Australia. ACM, New York, NY, USA, 9 pages. https://doi.org/10.1145/3664647.3680866

About

Official Implementation of [ACM-MM'24] Effective Optimization of Root Selection Towards Improved Explanation of Deep Classifiers

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages