ExplainabilityTransparencyAccountabilitycs.CV
ShapBPT: Image Feature Attributions Using Data-Aware Binary Partition Trees
Muhammad Rashid, Elvio G. Amparore, Enrico Ferrari, Damiano Verda
February 4, 2026
AAAI Conference on Artificial Intelligence
4 citations
1 influential
Key Takeaways
-
Better alignment with how images actually work: ShapBPT improves upon existing explanation methods by using a hierarchical structure (Binary Partition Trees) that respects the natural multiscale organization of images, meaning the explanations better match actual visual features like objects and regions rather than arbitrary pixel groupings.
-
Faster and more efficient explanations: The method demonstrates improved computational efficiency compared to existing explainability approaches, which is practically important since faster explanations make it more feasible to audit and validate AI systems used in real-world applications.
-
More meaningful visual insights for stakeholders: A 20-subject user study validated that ShapBPT provides more semantically meaningful explanations, suggesting that non-technical stakeholders and domain experts can better understand and trust how computer vision models make decisions when using this approach.
-
Addresses a gap in structured data interpretation: The paper identifies and fills an important gap—previous Shapley-based explanation methods hadn't been designed specifically for the structured nature of image data, meaning this work makes a foundational contribution to explaining vision models more appropriately than treating images like unstructured data.
Abstract
Pixel-level feature attributions are an important tool in eXplainable AI for Computer Vision (XCV), providing visual insights into how image features influence model predictions. The Owen formula for hierarchical Shapley values has been widely used to interpret machine learning (ML) models and their learned representations. However, existing hierarchical Shapley approaches do not exploit the multiscale structure of image data, leading to slow convergence and weak alignment with the actual morphological features. Moreover, no prior Shapley method has leveraged data-aware hierarchies for Computer Vision tasks, leaving a gap in model interpretability of structured visual data. To address this, this paper introduces ShapBPT, a novel data-aware XCV method based on the hierarchical Shapley formula. ShapBPT assigns Shapley coefficients to a multiscale hierarchical structure tailored for images, the Binary Partition Tree (BPT). By using this data-aware hierarchical partitioning, ShapBPT ensures that feature attributions align with intrinsic image morphology, effectively prioritizing relevant regions while reducing computational overhead. This advancement connects hierarchical Shapley methods with image data, providing a more efficient and semantically meaningful approach to visual interpretability. Experimental results confirm ShapBPT's effectiveness, demonstrating superior alignment with image structures and improved efficiency over existing XCV methods, and a 20-subject user study confirming that ShapBPT explanations are preferred by humans.