CoPS: Conditional Prompt Synthesis for Zero-Shot Anomaly Detection
Qiyu Chen, Zhen Qu, Wei Luo, Haiming Yao, Yunkang Cao, Yuxin Jiang, Yinan Duan,
Huiyuan Luo, Chengkan Lv*, Zhengtao Zhang
CoPS dynamically synthesizes visually conditioned prompts to fine‑tune CLIP, achieving SOTA zero‑shot anomaly detection. The source code and model checkpoints for CoPS will be released upon the paper’s acceptance. Stay tuned. Here, we present a brief summary of CoPS's performance across 5 industrial and 8 medical datasets:
| Metrics | MVTec-AD | VisA | BTAD | MPDD | DTD-Synthetic | HeadCT | BrainMRI | Br35H | ISIC | CVC-ColonDB | CVC-ClinicDB | Kvasir | Endo |
| I-AUROC (%) | 95.0 | 85.4 | 93.6 | 78.6 | 95.2 | 96.1 | 97.4 | 98.7 | – | – | – | – | – |
| P-AUROC (%) | 93.4 | 95.7 | 94.6 | 97.5 | 98.4 | – | – | – | 93.8 | 85.6 | 88.8 | 85.8 | 90.0 |