Research map / Vision-language and multimodal / CLIP and zero-shot
VCP-CLIP: A visual context prompting model for zero-shot anomaly segmentation
Zhen Qu, Xian Tao, Mukesh Prasad, Fei Shen, Zhengtao Zhang, Xinyi Gong, Guiguang Ding
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 Post-VCP and Pre-VCP on top of the DTP baseline (ablation) | VCP-CLIP · MVTec-AD | pixel AP: 45.2 -> 49.4 | Improved |
| add Post-VCP (fine-grained visual context) (ablation) | VCP-CLIP · MVTec-AD | pixel AP: 46.1 -> 49.4 | Improved |
| add Pre-VCP (global visual context in prompt) (ablation) | VCP-CLIP · MVTec-AD | pixel AP: 47.5 -> 49.4 | Improved |
| add deep text prompting (learnable embeddings in every text layer) (ablation) | VCP-CLIP · MVTec-AD | pixel AP: 48.2 -> 49.4 | Improved |
| ensemble patch features from layers 6,12,18,24 instead of a single layer (ablation) | VCP-CLIP · MVTec-AD | pixel AP: 44.1 -> 49.4 | Improved |
| increase input resolution beyond pretraining 336 to 798 (ablation) | VCP-CLIP · MVTec-AD | pixel AP: 49.4 -> 38.4 | Hurt |
| use ViT-L-14-336 instead of ViT-B-16-224 backbone (ablation) | VCP-CLIP · MVTec-AD | pixel AP: 37.9 -> 49.4 | Improved |
| vary state words in text prompt (good/damaged vs others) (ablation) | VCP-CLIP · MVTec-AD | pixel AP: 48.5 to 49.4 across state words | No clear effect |
| add VCP so prompts adapt to image rather than unified learned prompt | baseline prompts · MVTec-AD and VisA, state words changed at test time | AP | Improved |
Later papers that compared with this method
2 later accepted papers report VCP-CLIP as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- From Attraction to Equilibrium: Physics-Inspired Semantic Gravitons for Zero-Shot Anomaly Detection (CVPR 2026 · better on 1 of 1 shared result)
- MuSc-V2: Zero-Shot Multimodal Industrial Anomaly Classification and Segmentation With Mutual Scoring of Unlabeled Samples (TPAMI 2026 · better on 1 of 1 shared result)
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.
- WinCLIP — WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentation (CVPR 2023) · better on 1 of 1 shared result
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
- AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection (ICLR 2024 · similarity 0.89)
- Aligning and Prompting Anything for Zero-Shot Generalized Anomaly Detection (AAAI 2025 · similarity 0.89)
- Bayesian Prompt Flow Learning for Zero-Shot Anomaly Detection (CVPR 2025 · similarity 0.88)
- AA-CLIP: Enhancing Zero-Shot Anomaly Detection via Anomaly-Aware CLIP (CVPR 2025 · similarity 0.88)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.86)
- MetaUAS: Universal Anomaly Segmentation with One-Prompt Meta-Learning (NeurIPS 2024 · similarity 0.86)
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.
