Research map / Video / Video-language understanding
Aligning Effective Tokens with Video Anomaly in Large Language Models
Yingxian Chen, Jiahui Liu, Ruidi Fan, Yanwei Li, Chirui Chang, Shizhen Zhao, Wilton W. T. Fok, Xiaojuan Qi, Yik-Chung Wu
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 Spatial Effective Token Selection (SETS) without fine-tuning (ablation) | LLaMA-VID baseline · UCF-Crime | Total Acc.: 14.83 -> 24.83 | Improved |
| add Temporal Effective Tokens (TETG) without fine-tuning (ablation) | LLaMA-VID baseline · UCF-Crime | Total Acc.: 14.83 -> 23.79 | Improved |
| add TETG on top of SETS, no fine-tuning (ablation) | VA-GPT · UCF-Crime | Total Acc.: 24.83 -> 25.12 | Improved |
| add TETG on top of SETS, stage two (ablation) | VA-GPT · UCF-Crime | Temporal Acc.: 31.60 -> 35.00 | Improved |
| add anomaly video fine-tuning (stage one) (ablation) | VA-GPT · UCF-Crime | Total Acc.: 25.12 -> 27.50 | Improved |
| add stage two fine-tuning with SETS (ablation) | VA-GPT · UCF-Crime | Total Acc.: 27.50 -> 30.69 | Improved |
| fine-tune on UCF instruct-following data | LLaMA-VID · UCF-Crime | Total Acc.: 14.83 -> 23.10 | Improved |
| fine-tune on UCF instruct-following data | Video-ChatGPT · UCF-Crime | Total Acc.: 24.13 -> 26.23 | Improved |
| fine-tune on UCF instruct-following data | VA-GPT · UCF-Crime | Total Acc.: 25.12 -> 30.69 | Improved |
| raise SETS sample rate K from 0.5 to 0.9 (ablation) | VA-GPT · UCF-Crime | Total Acc.: 30.69 -> 27.27 | Hurt |
| lower SETS sample rate K from 0.5 to 0.1 (ablation) | VA-GPT · UCF-Crime | Total Acc.: 30.69 -> 23.61 | Hurt |
| raise SETS sample rate K from 0.5 to 0.7 (ablation) | VA-GPT · UCF-Crime | Temporal Acc.: 35.00 -> 31.23 | Hurt |
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.
- LLaMA-VID — LLaMA-VID: An Image is Worth 2 Tokens in Large Language Models (ECCV 2024) · better on 4 of 4 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
- Harnessing Large Language Models for Training-free Video Anomaly Detection (CVPR 2024 · similarity 0.91)
- Holmes-VAU: Towards Long-term Video Anomaly Understanding at Any Granularity (CVPR 2025 · similarity 0.89)
- MoniTor: Exploiting Large Language Models with Instruction for Online Video Anomaly Detection (NeurIPS 2025 · similarity 0.88)
- Ex-VAD: Explainable Fine-grained Video Anomaly Detection Based on Visual-Language Models (ICML 2025 · similarity 0.88)
- E.T. Bench: Towards Open-Ended Event-Level Video-Language Understanding (NeurIPS 2024 · similarity 0.88)
- Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · similarity 0.87)
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.
