Research map / Label-efficient and robust learning / OOD and anomaly detection
Neural Batch Sampling with Reinforcement Learning for Semi-Supervised Anomaly Detection
Wen-Hsuan Chu, Kris M. Kitani
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 |
|---|---|---|---|
| use loss-profile classifier with RL batch sampler instead of U-Net trained on 5 labeled images | semi-supervised anomaly segmentation · MVTec AD Bottle | F1: 0.31 -> 0.80 | Improved |
| use loss-profile classifier with RL batch sampler instead of U-Net | semi-supervised anomaly segmentation · MVTec AD Metal Nut | F1: 0.29 -> 0.82 | Improved |
| use loss-profile classifier with RL batch sampler instead of U-Net | semi-supervised anomaly segmentation · CrackForest | F1: 0.21 -> 0.36 | Improved |
| use loss-profile classifier with RL batch sampler instead of U-Net | semi-supervised anomaly segmentation · NanoTWICE | F1: 0.45 -> 0.33 | Hurt |
| use loss-profile classifier with RL batch sampler instead of U-Net | semi-supervised anomaly segmentation · NanoTWICE | recall: 0.59 -> 0.80 | Improved |
| use proposed method instead of unsupervised reconstruction-loss thresholding | semi-supervised anomaly segmentation · MVTec AD Bottle | F1: 0.34 -> 0.80 | Improved |
| use U-Net with 5 labeled images instead of unsupervised thresholding | U-Net · MVTec AD Carpet | F1: 0.08 -> 0.43 | Improved |
| use proposed method instead of U-Net on small-object class | semi-supervised anomaly segmentation · MVTec AD Screw | F1: 0.01 -> 0.08 | No clear effect |
| use proposed method instead of U-Net on small-object class | semi-supervised anomaly segmentation · MVTec AD Capsule | F1: 0.05 -> 0.12 | No clear effect |
| normalize rewards in REINFORCE with a single terminal reward | neural batch sampler · MVTec AD | training stability | Hurt |
| train sampler with sparse reward only instead of adding behavior cloning bootstrapping | neural batch sampler · anomaly detection | training efficiency | Improved |
Closest papers by meaning
- Few-Shot Anomaly-Driven Generation for Anomaly Classification and Segmentation (ECCV 2024 · similarity 0.87)
- Two-Layer Convolutional Autoencoders Trained on Normal Data Provably Detect Unseen Anomalies (ICLR 2026 · similarity 0.87)
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · similarity 0.85)
- Toward Generalist Anomaly Detection via In-context Residual Learning with Few-shot Sample Prompts (CVPR 2024 · similarity 0.85)
- One-for-More: Continual Diffusion Model for Anomaly Detection (CVPR 2025 · similarity 0.85)
- Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.85)
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.
