ShapBPT: Image Feature Attributions Using Data-Aware Binary Partition Trees
Muhammad Rashid, Elvio G. Amparore, Enrico Ferrari, Damiano Verda · 2026
Read ShapBPT: Image Feature Attributions Using Data-Aware Binary Partition Trees by Muhammad Rashid, Elvio G. Amparore, Enrico Ferrari, Damiano Verda (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07047.txt.
Description
Muhammad Rashid and three co-authors explain image model decisions at pixel level with a new method, ShapBPT. Pixel-level attributions help users see which image features sway a prediction. The Owen formula for hierarchical Shapley values is widely used for such explanations, yet old hierarchical methods ignore how image data nests across scales. That leads to slow convergence and weak fit to true morphological features, and no prior Shapley method had used data-aware hierarchies for vision.
ShapBPT ties Shapley coefficients to a binary partition tree built for images, so attributions track intrinsic image morphology. Relevant regions gain priority, and compute overhead falls. Tests show better fit to image structures and better efficiency than old explainable vision methods. A study with 20 human subjects found people preferred ShapBPT explanations, which links hierarchical Shapley theory to structured visual data in a more efficient, meaningful way.
Catalog record
| CREATOR | Muhammad Rashid, Elvio G. Amparore, Enrico Ferrari, Damiano Verda |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07047