BSED: Baseline Shapley-Based Explainable Detector

Shapley values provide mathematically rigorous explanations: BSED uses Shapley values to explain object detection predictions while satisfying formal "explainability axioms"—mathematical guarantees that explanations are valid and trustworthy, addressing a key problem where some existing explanation methods produce irrelevant or misleading information.

Practical computational efficiency enables real-world deployment: While traditional Shapley value calculations are prohibitively expensive, BSED achieves reasonable processing costs, making mathematically sound AI explanations feasible for actual use rather than remaining a theoretical ideal.

Model-agnostic approach increases accessibility: BSED works across different detection systems without requiring specialized tuning for each model, meaning organizations can apply rigorous explainability standards to their existing AI tools rather than needing to rebuild systems from scratch.

Explanations can improve AI system performance: The paper demonstrates practical applications where explanations from BSED can be used to correct detection errors, suggesting that understanding why an AI makes decisions can help humans identify and fix problems in the system's reasoning.

Addresses a critical gap in visual AI transparency: While explainability methods have advanced for image classification, object detection—which identifies and locates multiple items in images—had lagged behind; this work extends rigorous explanation methods to this widely-used AI capability.