Research map / Label-efficient and robust learning / Recognition and long-tail
UniSpector: Towards Universal Open-set Defect Recognition via Spectral-Contrastive Visual Prompting
Geonuk Kim, Minhoi Kim, Kangil Lee, Minsu Kim, Hyeonseong Jeon, Jeonghoon Han, Hyoungjoon Lim, Junho Yim
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-Spectral Prompt Encoder (SSPE) (ablation) | visual prompting baseline · Real-IAD unseen | AP50 box: 24.0 -> 43.0 | Improved |
| add Contrastive Prompt Encoder (CPE) on top of SSPE (ablation) | UniSpector · Real-IAD unseen | AP50 box: 43.0 -> 65.8 | Improved |
| add Prompt-guided Query Selection (PQS) (ablation) | UniSpector (SSPE+CPE) · Real-IAD unseen | AP50 box: 65.8 -> 69.1 | Improved |
| replace learnable queries with heuristic top-K similarity selection (ablation) | UniSpector · GC10 | AP50 box: 34.4 -> 35.6 | No clear effect |
| replace heuristic top-K with differentiable PQS (ablation) | UniSpector · VisA | AP50 box: 30.1 -> 32.8 | Improved |
| fine-tune on InsA in-domain seen sets | DINOv · InsA average | AP50 box: 10.2 -> 17.1 | Improved |
| fine-tune on InsA in-domain seen sets | YOLOE · InsA average | AP50 box: 17.1 -> 17.4 | No clear effect |
| use visual prompting DINOv and T-Rex2 in closed-set Real-IAD | naive prompt-to-region matching · Real-IAD closed-set | AP50 box: DINOv 24.8, T-Rex2 29.8 vs UniSpector 90.0 | Hurt |
| use text prompts (GroundingDINO, YOLO-World) instead of visual prompts | open-set defect localization · InsA cross-domain | AP50 box: 3CAD 0.0 for GroundingDINO | Hurt |
Closest papers by meaning
- Towards Open-Vocabulary Industrial Defect Understanding with a Large-Scale Multimodal Dataset (CVPR 2026 · similarity 0.85)
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024 · similarity 0.85)
- Zero-Shot Learning in Industrial Scenarios: New Large-Scale Benchmark, Challenges and Baseline (AAAI 2025 · similarity 0.84)
- PromptAD: Learning Prompts with only Normal Samples for Few-Shot Anomaly Detection (CVPR 2024 · similarity 0.84)
- MVREC: A General Few-shot Defect Classification Model Using Multi-View Region-Context (AAAI 2025 · similarity 0.83)
- Unsupervised Continual Anomaly Detection with Contrastively-Learned Prompt (AAAI 2024 · similarity 0.83)
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.
