Research map / Vision-language and multimodal / CLIP and zero-shot
VisualAD: Language-Free Zero-Shot Anomaly Detection via Vision Transformer
Yanning Hou, Peiyuan Li, Zirui Liu, Yitong Wang, Yanran Ruan, Jianfeng Qiu, Ke Xu
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 SCA module (ablation) | VisualAD (CLIP) · VisA (image-level) | AUROC: 82.3 -> 84.7 | Improved |
| add SAF module (ablation) | VisualAD (CLIP) · VisA (image-level) | AUROC: 50.5 -> 84.7 | Improved |
| add SAF module (ablation) | VisualAD (CLIP) · VisA (pixel-level) | AP: 3.5 -> 28.4 | Improved |
| add both SCA and SAF (ablation) | VisualAD (CLIP) · VisA (image-level) | AUROC: 48.0 -> 84.7 | Improved |
| add focal loss (ablation) | VisualAD (CLIP) · VisA (image-level) | AUROC: 83.5 -> 84.7 | Improved |
| add dice loss (ablation) | VisualAD (CLIP) · VisA (image-level) | AUROC: 83.4 -> 84.7 | Improved |
| add token contrastive loss (ablation) | VisualAD (CLIP) · VisA (image-level) | AUROC: 84.2 -> 84.7 | Improved |
| increase anchors from 4 to 16 (ablation) | VisualAD (CLIP) · VisA (image-level) | AUROC: 84.7 -> 84.9 | No clear effect |
| increase anchors from 4 to 8 (ablation) | VisualAD (CLIP) · VisA (pixel-level) | AUROC: 95.8 -> 95.0 | Hurt |
| use all four layers instead of only layer 18 (ablation) | VisualAD (CLIP) · VisA (image-level) | AUROC: 82.4 -> 84.7 | Improved |
| add layer 24 to layers {6,12,18} (ablation) | VisualAD (CLIP) · VisA (image-level) | AUROC: 83.4 -> 84.7 | Improved |
| add positional encoding in SCA (ablation) | VisualAD (CLIP) · VisA (pixel-level) | AUROC: 93.0 -> 95.8 | Improved |
Closest papers by meaning
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.94)
- AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection (ICLR 2024 · similarity 0.91)
- Aligning and Prompting Anything for Zero-Shot Generalized Anomaly Detection (AAAI 2025 · similarity 0.90)
- UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection (CVPR 2025 · similarity 0.90)
- Adapting Visual-Language Models for Generalizable Anomaly Detection in Medical Images (CVPR 2024 · similarity 0.89)
- AA-CLIP: Enhancing Zero-Shot Anomaly Detection via Anomaly-Aware CLIP (CVPR 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.
