Research map / Detection and segmentation / Open-set and few-shot detection
Learning Feature Inversion for Multi-class Anomaly Detection under General-purpose COCO-AD Benchmark
Jiangning Zhang, Chengjie Wang, Xiangtai Li, Guanzhong Tian, Zhucun Xue, Yong Liu, Guansong Pang, Dacheng Tao
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 skip autoencoder structure to RD | RD · MVTec AD | mIoU-max: 37.3 -> 41.7 | Improved |
| add rescaled features phi^R | RD + skip AE · MVTec AD | mAD_I: 96.5 -> 97.7 | Improved |
| add style features phi^S (ablation) | InvAD · MVTec AD | mIoU-max: 41.8 -> 42.5 | Improved |
| add SSM modulation in place of concatenation and convolution (ablation) | InvAD · MVTec AD | mIoU-max: 41.8 -> 43.0 | Improved |
| use both style features and SSM (ablation) | InvAD · MVTec AD | mAD_P: 70.3 -> 72.0 | Improved |
| widen channels N_R/N_S from 64/16 to 256/64 (ablation) | InvAD · MVTec AD | mAD^.2_.8: 34.8 -> 34.0 | No clear effect |
| shrink channels to 16/8 (ablation) | InvAD · MVTec AD | mAD_P: 72.0 -> 71.1 | No clear effect |
| increase stack to 3/3/4 from 1/1/2 (ablation) | InvAD · MVTec AD | mIoU-max: 43.7 -> 42.7 | Hurt |
| use finer metric step 0.01 instead of 0.1 (ablation) | M^.2_.8 metrics · MVTec AD | mF1-max^.2_.8 and time: 34.58 -> 37.02; 13.59 s -> 115.74 s | Mixed |
| widen threshold range to 0.1-0.9 (ablation) | M^.2_.8 metrics · MVTec AD | mF1-max^.2_.8: 34.58 -> 28.54 | Hurt |
| narrow threshold range to 0.4-0.6 (ablation) | M^.2_.8 metrics · MVTec AD | mF1-max^.2_.8: 34.58 -> 44.06 | Improved |
| replace WideResNet-50 with ResNet34 (InvAD-lite) | InvAD · MVTec AD | mAD_P: 72.0 -> 70.2 | Mixed |
Closest papers by meaning
- ADPretrain: Advancing Industrial Anomaly Detection via Anomaly Representation Pretraining (NeurIPS 2025 · similarity 0.88)
- ResAD: A Simple Framework for Class Generalizable Anomaly Detection (NeurIPS 2024 · similarity 0.88)
- A Diffusion-Based Framework for Multi-Class Anomaly Detection (AAAI 2024 · similarity 0.87)
- RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection (CVPR 2024 · similarity 0.87)
- Long-Tailed Anomaly Detection with Learnable Class Names (CVPR 2024 · similarity 0.87)
- AnomalyNCD: Towards Novel Anomaly Class Discovery in Industrial Scenarios (CVPR 2025 · similarity 0.86)
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.
