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.
Figures & tables
Figure 1: LLM/MLLM-as-a-judge vs. claim-based verification. (a) A judge takes the ground truth, evaluation rubric, and explanation as input and autoregressively generates a score. (b) AnomalyJev predicts one option per ground-truth attribute directly from contextualised representations, then verifies the resulting claims to produce deterministic scores, without token generation.
Figure 2: Overview of AnomalyJev. Training (orange): VLM explanations are augmented and weakly labelled by an LLM to supervise the LoRA module and query specific pointer heads. Inference (blue): the explanation is encoded once, question branches are processed in parallel, and the pointer head selects one option from each question’s option set without token generation. The extracted claims are verified against the ground truth to score faithfulness and completeness.
Figure 3: Agreement with human experts (Spearman ρ , faithfulness) against model size and runtime.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 4: Screenshot of the interface used by human raters.
Figure 5: Instructions given to the two experts for the faithfulness (“correctness”) rating.
Figure 6: Instructions given to the two experts for the completeness rating.
Multimodal Large Language Models (MLLMs) provide a natural way to make video anomaly detection more explainable. However, their final decisions do not always fully use the discriminative information contained in their hidden states, an issue we refer to as representation--behavior misalignment. We decompose this gap into a capacity component that measures discriminative information never aggregated into the readout position, and a directional component that measures the angular mismatch between the optimal and the native normal--abnormal axis at that position. Across multiple video anomaly detection benchmarks and MLLM backbones the directional component dominates, and residual-stream tracing shows that native-axis separability rises sharply in several mid-to-late attention layers. Because both components are governed by attention rather than MLP updates, we propose Representation--Behavior Alignment (RBA), a parameter-efficient method that adapts those layers using video-level labels alone while updating about 0.012% of the backbone parameters. Experiments on three benchmarks show that RBA improves native-readout performance and better aligns the model's decision direction with discriminative representations, and it produces anomaly decisions and explanations through a single generative process.
Vision-language models (VLMs) have recently emerged as a promising paradigm for video anomaly detection (VAD) due to their strong visual reasoning ability and natural language-based explainability. In this paper, we aim to address a key limitation of such pipelines, which perform segment-level inference independently owing to token constraints and reason without structured temporal context, allowing VLMs to interpret anomalies as deviations from evolving video dynamics rather than producing fragmented predictions and explanations. To specify, we propose a context-aware framework named LATERN, which reformulates VAD as a temporal evidence aggregation process. LATERN consists of two complementary modules: Context-Aware Anomaly Scoring (CEA) and Recursive Evidence Aggregation (REA). CEA introduces a novel image-grounded memory mechanism, which selectively chooses historical content via frame diversity and visual-textual alignment as expanded context to help generate reliable anomaly scores. Building upon these scores, REA performs recursive temporal aggregation to identify coherent anomaly intervals and produce event-level decisions and explanations grounded in visual-textual evidence. Extensive experiments on challenging benchmarks, including UCF-Crime and XD-Violence, show that LATERN enhances detection accuracy and explanation consistency for frozen VLMs during test time, while generating temporally coherent and semantically grounded event-level explanations.
Anomaly detection aims to identify observations that deviate from normal patterns. Recent work has increasingly used pretrained vision--language models (VLMs) for training-free image and video anomaly detection without task-specific retraining. Anomaly detection is commonly evaluated by how well anomaly scores rank anomalous images or video frames above normal ones. Generative VLMs, however, assign probabilities to possible answers and then decode a single answer. This decoding step may discard ordering information. We call this loss decoded-answer rank compression. To isolate this effect, we compare two ways of scoring the same VLM output: one uses only the decoded answer, while the other computes a probability-weighted score over all possible answers. Probability-weighted scoring consistently outperforms decoded-answer scoring across image and video anomaly detection benchmarks, using different VLMs and answer scales. The mean gains range from 7.66 to 19.95 points on the primary metrics. Most of this gap comes from decoded-answer ties. Breaking decoded-answer ties with answer probabilities recovers at least 95% of the average performance gap on every benchmark. We further investigate how these ties affect evaluation. We find that evaluating tied scores one input at a time makes the reported AUROC depend on input order, and that linear PR interpolation can inflate reported performance. These results reveal that both how VLM outputs are scored and how discrete anomaly scores are evaluated affect reported anomaly-detection performance.