Evaluation method improves checking explanations of image anomaly detectors
AnomalyJev: Towards Evaluating Image Anomaly Explanations via Claim-Level Verification
Computer Vision and Pattern RecognitionArtificial IntelligenceComputation and Language
Summary
Detecting unusual things in images is helpful, but understanding why a computer thinks something is unusual can be tricky. The authors found that existing ways to check if explanations make sense are slow and costly because they use lots of computer language generation. Instead, they developed a simpler way to pull out checkable statements from explanations and compare them to known facts. Their method, called AnomalyJev, matches human judgment closely while running faster and using fewer resources.
What this means in practice
- •For machine learning engineers: Improve the reliability and efficiency of evaluating image anomaly explanations during model development by using claim-based verification.
- •For computer vision teams: Use a lightweight framework to verify textual explanations from anomaly detectors against annotations, speeding up quality control of detection systems.
Authors
Qizhou Wang, Jiawen Zhu, Guansong Pang, Christopher Leckie
Abstract
Recent multimodal LLM (MLLM)-based anomaly detection (AD) methods can generate textual explanations that help interpret their predictions. However, reliably and efficiently evaluating these explanations remains challenging. While LLM judges can assess explanations through structured rubrics, they often require substantial token generation, making evaluation costly and difficult to scale. We instead propose to anchor evaluation in the available ground truth by extracting verifiable claims from each explanation and verifying them against the corresponding annotations. We instantiate this idea as AnomalyJev, a lightweight framework that extracts verifiable claims from contextualised representations using ground-truth-defined queries and pointer based heads, without token generation. The claims are then verified against the ground truth to score faithfulness and completeness. For evaluation, we construct a human-annotated set of 640 explanations with over 1,600 expert ratings across in-domain and cross-dataset settings. Despite its compact design, AnomalyJev achieves stronger agreement with human annotators than most open MLLM judges and approaches that of flagship MLLM judges, at a fraction of their runtime and resources.