AI Paper Atlas

Research map / Video / Video-language understanding

Towards Trustworthy Video Anomaly Understanding: A Class-Guided Chain-of-Evaluation Metric and An Anomaly-focused Meta-Benchmark

Jiaxu Leng, Zhoujie Huang, Mingpi Tan, Zhanjie Wu, Xinbo Gao

ICML 2026 · Video › Video-language understanding · approach: Event and anomaly reasoning in video

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add anomalous event extraction and uniform rewriting (ablation)CG-CoE (P) · AEArho_Prec: 0.682 -> 0.727Improved
add anomalous event extraction and uniform rewriting (ablation)CG-CoE (P) · CVPmean score variation (lower is better): 0.118 -> 0.072Improved
add compositional matching (ablation)CG-CoE (P) · AEArho_Prec: 0.727 -> 0.772Improved
add compositional matching (ablation)CG-CoE (P) · CVPmean score variation (lower is better): 0.072 -> 0.063Improved
add class-guided tolerance boundary (ablation)CG-CoE (P) · AEArho_Prec: 0.772 -> 0.800Improved
add class-guided tolerance boundary (ablation)CG-CoE (R) · AEArho_Rec: 0.782 -> 0.813Improved
add class-guided tolerance boundary (ablation)CG-CoE (R) · CVPmean score variation (lower is better): 0.061 -> 0.051Improved
replace best LLM-based baseline (VAD-R1 classification) with CG-CoEmetric · AEArho_Prec: 0.731 -> 0.800Improved
replace HAWK (consistency) with CG-CoEmetric · CVPmean score variation (lower is better): 0.099 -> 0.059Improved
replace BARTScore with CG-CoEmetric · AEA vs CVPrho_Prec / variation: 0.313 -> 0.800; 0.008 -> 0.059Mixed
use Qwen3.5-9B or larger judge instead of 4B or smaller (ablation)CG-CoE · LLM scale, 100 samplesevaluation reliability: N/A -> 0.861Improved
scale judge LLM from Qwen3.5-9B to 197B (ablation)CG-CoE · LLM scale, CVPmean score variation (lower is better): 0.121 -> 0.070Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy