Research map / Label-efficient and robust learning / OOD and anomaly detection
HLRAD: High-dimensional Latent Representation for Unified Anomaly Detection
Tianrui Zhang, Ziheng Zang, Ningmu Zou
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 all four components (Senc, SGF, DM, HD) to reconstruction baseline (ablation) | baseline · MVTec-AD (multi-class) | P-AP: 69.3 -> 71.3 | Improved |
| add all four components to baseline (ablation) | baseline · VisA (multi-class) | P-AP: 53.2 -> 55.5 | Improved |
| add all four components to baseline (ablation) | baseline · MVTec-AD (multi-class) | I-F1max: 99.0 -> 99.3 | Improved |
| widen decoder from 768 to 2048 (ablation) | HLRAD · MVTec-AD (multi-class) | P-AP: 70.0 -> 71.3 | Improved |
| deepen decoder from 4 to 8 blocks (ablation) | HLRAD · MVTec-AD (multi-class) | P-AP: 71.0 -> 71.3 | Improved |
| deepen decoder from 8 to 10 blocks (ablation) | HLRAD · MVTec-AD (multi-class) | P-AP: 71.3 -> 71.3 | No clear effect |
| freeze semantic encoder (ablation) | DiT semantic encoder · MVTec-AD (multi-class) | P-AP: 70.9 -> 71.3 | Improved |
| freeze semantic encoder (ablation) | CLIP semantic encoder · MVTec-AD (multi-class) | P-AP: 69.5 -> 69.8 | Improved |
| use DiT instead of CLIP as frozen semantic encoder (ablation) | HLRAD · MVTec-AD (multi-class) | P-AP: 69.8 -> 71.3 | Improved |
| use DINOv3-L instead of DINOv2-L encoder (ablation) | HLRAD · MVTec-AD (multi-class) | P-AP: 69.8 -> 71.3 | Improved |
| use DINOv3-L instead of DINOv2-L encoder (ablation) | native baseline · MVTec-AD (multi-class) | I-AUROC: 99.5 -> 99.6 | No clear effect |
| add HLRAD framework to native reconstruction baseline (ablation) | DINOv3-L · MVTec-AD (multi-class) | P-AP: 70.4 -> 71.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Dinomaly — Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ResAD: A Simple Framework for Class Generalizable Anomaly Detection (NeurIPS 2024 · similarity 0.87)
- Hyperbolic Anomaly Detection (CVPR 2024 · similarity 0.87)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.87)
- Long-Tailed Anomaly Detection with Learnable Class Names (CVPR 2024 · similarity 0.86)
- MambaAD: Exploring State Space Models for Multi-class Unsupervised Anomaly Detection (NeurIPS 2024 · similarity 0.86)
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (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.
