🧪 Test?View on arXiv
Confidence-Controlled XAI Auditing for Pedestrian Detection under Domain Shift
Not provided in the abstract
explainabilitypedestrian detectiondomain shiftXAI auditing
2610.02364
Builder Relevance
1h ago70%
Abstract
This paper audits the faithfulness of post-hoc explanations for a pedestrian detection model under domain shift conditions.
Reality Card
Core Claim
The study demonstrates that deletion-based faithfulness of explanations is strongly linked to detection strength, revealing that naive confidence-stratified comparisons are unreliable.
Method / Result
Spearman correlations between 0.70 and 0.82 for D-RISE indicate strong coupling between deletion-based faithfulness and detection strength.
Limitations
The results are domain-dependent, which may limit generalizability across different datasets.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.
No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.