Research map / Video / Video-language understanding
CueBench: Advancing Unified Understanding of Context-Aware Video Anomalies in Real-World
Yating Yu, Congqi Cao, Zhaoying Wang, Weihua Meng, Jie Li, Yuxin Li, Zihao Wei, Zhongpei Shen, Jiajun Zhang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add SFT then RFT instead of baseline Qwen2.5-VL-3B | CUE-R1 · CueBench event recognition | semantic score: 35.5 -> 73.2 | Improved |
| add RFT after SFT (ablation) | CUE-R1 · CueBench anomaly recognition top-down | hierarchy score: 7.1 -> 7.7 | Improved |
| add RFT after SFT (ablation) | CUE-R1 · CueBench anomaly recognition bottom-up | hierarchy score: 8.1 -> 13.6 | Improved |
| add RFT after SFT (ablation) | CUE-R1 · CueBench temporal grounding | temporal IoU: 34.6 -> 35.9 | Improved |
| add RFT after SFT (ablation) | CUE-R1 · CueBench event recognition | semantic score: 73.0 -> 73.2 | Mixed |
| add SFT before RFT (ablation) | CUE-R1 · CueBench anomaly detection | temporal IoU: RFT only 34.9, SFT+RFT 35.2 | No clear effect |
| add SFT only to Qwen2.5-VL-3B (ablation) | Qwen2.5-VL-3B · CueBench event recognition | semantic score: 35.5 -> 73.0 | Improved |
| add RFT only to Qwen2.5-VL-3B (ablation) | Qwen2.5-VL-3B · CueBench anomaly recognition top-down | hierarchy score: 1.5 -> 3.1 | No clear effect |
| add RFT only to Qwen2.5-VL-3B (ablation) | Qwen2.5-VL-3B · CueBench anticipation | hierarchy score: 0.4 -> 0.6 | No clear effect |
| apply R1-style RL to a video VLM (Video-R1) | Video-R1 · CueBench event recognition | semantic score: 9.5 vs Qwen2.5-VL-3B 35.5 | Hurt |
| use a proprietary model (Gemini-1.5-flash) instead of open-source baselines | Gemini-1.5-flash · CueBench scene recognition | semantic score: 59.9 vs Qwen2.5-VL-3B 41.7 | Improved |
| use specialized VAU models instead of general action recognition | Holmes-VAU · CueBench anomaly recognition (specialized) | Top-1: 10.72 -> 29.72 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Holmes-VAU — Holmes-VAU: Towards Long-term Video Anomaly Understanding at Any Granularity (CVPR 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Holmes-VAU: Towards Long-term Video Anomaly Understanding at Any Granularity (CVPR 2025 · similarity 0.91)
- HAWK: Learning to Understand Open-World Video Anomalies (NeurIPS 2024 · similarity 0.89)
- Open-Vocabulary Video Anomaly Detection (CVPR 2024 · similarity 0.89)
- VERA: Explainable Video Anomaly Detection via Verbalized Learning of Vision-Language Models (CVPR 2025 · similarity 0.89)
- Harnessing Large Language Models for Training-free Video Anomaly Detection (CVPR 2024 · similarity 0.88)
- VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Tree (NeurIPS 2025 · similarity 0.88)
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.
