Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
FuDU: A Fuzzy Dual-dimension Uncertainty Framework for Streaming Active Learning in Industrial Defect Detection
Zhaoyang Wang, Haiyong Chen, Binyi Su, Xinwei Lyu
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 |
|---|---|---|---|
| replace confidence sampling with FuDU fuzzy dual-uncertainty sampling | RT-DETR streaming AL · Nuclear fuel rod defects | mean mAP50: 87.2 -> 92.2 | Improved |
| use FuDU instead of confidence across backbones and detection heads | Swin, ViTDet, DEIM, MR-DETR, MI-DETR · Nuclear fuel rod defects | mean mAP50: 87.3 -> 91.6 (Swin); 86.6 -> 93.2 (MI-DETR) | Improved |
| set prototype count k to 50 rather than 10 (ablation) | PGUQ · Nuclear fuel rod defects | sampling ratio: 40.5% -> 15.4% with mAP 92.0 -> 92.2 | Improved |
| set prototype count k to 50 rather than 100 (ablation) | PGUQ · Nuclear fuel rod defects | mean mAP50: 88.5 -> 92.2 | Improved |
| learnable end-to-end prototypes instead of re-clustering each round (ablation) | PGUQ · Nuclear fuel rod defects | mean mAP50: 90.8 -> 92.2 | Improved |
| prototype distance instead of reconstruction-based global uncertainty (ablation) | PGUQ · Nuclear fuel rod defects | mean mAP50: 89.7 (CD), 88.8 (FAE) -> 92.2 | Improved |
| add defect uncertainty Ud to global uncertainty (ablation) | FuDU · Nuclear fuel rod defects | mean mAP50: 89.6 -> 92.2 | Improved |
| add localization entropy weight (w2=2) with classification entropy (ablation) | DeUE · Nuclear fuel rod defects | mean mAP50: 90.2 -> 92.2 | Improved |
| increase classification entropy weight to 2 (ablation) | DeUE · Nuclear fuel rod defects | mean mAP50: 91.7 -> 89.8 | Hurt |
| increase localization weight to 4 (ablation) | DeUE · Nuclear fuel rod defects | sampling ratio: 15.4% -> 17.3% (mAP 92.2 -> 92.1) | Hurt |
| fuzzy inference sampling instead of hard threshold (ablation) | FuDU · Nuclear fuel rod defects | mean mAP50 / ratio: 92.0 -> 92.2; P 88.2 -> 95.1 | Improved |
| fuzzy inference instead of sigmoid mapping (ablation) | FuDU · Nuclear fuel rod defects | sampling ratio: 23.4% -> 15.4% | Improved |
Closest papers by meaning
- MAU-GPT: Enhancing Multi-type Industrial Anomaly Understanding via Anomaly-aware and Generalist Experts Adaptation (AAAI 2026 · similarity 0.83)
- Towards Open-Vocabulary Industrial Defect Understanding with a Large-Scale Multimodal Dataset (CVPR 2026 · similarity 0.81)
- InCLAD: A Continual Learning Benchmark for Industrial Visual Anomaly Detection (NeurIPS 2026 · similarity 0.80)
- MVREC: A General Few-shot Defect Classification Model Using Multi-View Region-Context (AAAI 2025 · similarity 0.80)
- Towards Training-free Anomaly Detection with Vision and Language Foundation Models (CVPR 2025 · similarity 0.80)
- Multi-Sensor Object Anomaly Detection: Unifying Appearance, Geometry, and Internal Properties (CVPR 2025 · similarity 0.80)
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.
