Research map / Video / Video segmentation and tracking
Alert-CLIP: Abnormality-aware Latent-Enhanced Representation Tuning of CLIP for Video Anomaly Detection
Yiyan Zhu, Menghao Zhang, Haifeng Sun, Pengfei Ren, Xianao Chu, Chenye Xu, Hong Tan, Jinghan Wang, Qi Qi, Jingyu Wang
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 region alignment loss (ablation) | Alert-CLIP global-only · UCF-Crime (weakly supervised) | AUC: 88.53 -> 88.85 | Improved |
| add semantic hard-negative loss (ablation) | Alert-CLIP global+region · UCF-Crime (weakly supervised) | AUC: 88.85 -> 89.32 | Improved |
| add global alignment only (ablation) | CLIP · UCF-Crime (weakly supervised) | AUC: 88.02 -> 88.53 | Improved |
| two-stage training instead of joint (ablation) | Alert-CLIP · UCF-Crime | AUC: 88.61 -> 89.32 | Improved |
| hard negatives instead of in-batch negatives (ablation) | Alert-CLIP · XD-Violence (zero-shot) | AP: 46.44 -> 55.18 | Improved |
| hard negatives instead of in-batch negatives (ablation) | Alert-CLIP · UCF-Crime (zero-shot) | AUC: 72.71 -> 75.75 | Improved |
| add stage 2 (region + hard negatives) after stage 1 (ablation) | Alert-CLIP · MSAD (weakly supervised) | AP: 75.10 -> 78.24 | Improved |
| replace CLIP features with Alert-CLIP stage 1 | VadCLIP · UCF-Crime (weakly supervised) | AP: 32.72 -> 32.16 | No clear effect |
| replace CLIP with Alert-CLIP (full) | zero-shot VAD · UCF-Crime | AUC: 61.91 -> 75.75 | Improved |
| replace CLIP with Alert-CLIP | ImageNet-1K classification · ImageNet-1K | Top-1: 68.34 -> 68.58 | No clear effect |
| replace CLIP with Alert-CLIP | Flickr30k retrieval · Flickr30k I2T | R@1: 82.20 -> 89.90 | Improved |
| replace OVVAD with Alert-CLIP | open-vocabulary VAD · UCF-Crime | AUC: 86.40 -> 86.14 | Hurt |
Closest papers by meaning
- VadCLIP: Adapting Vision-Language Models for Weakly Supervised Video Anomaly Detection (AAAI 2024 · similarity 0.90)
- Holmes-VAU: Towards Long-term Video Anomaly Understanding at Any Granularity (CVPR 2025 · similarity 0.88)
- Harnessing Large Language Models for Training-free Video Anomaly Detection (CVPR 2024 · similarity 0.87)
- Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · similarity 0.87)
- AA-CLIP: Enhancing Zero-Shot Anomaly Detection via Anomaly-Aware CLIP (CVPR 2025 · similarity 0.87)
- Prompt-Enhanced Multiple Instance Learning for Weakly Supervised Video Anomaly Detection (CVPR 2024 · 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.
