Anomaly Detection2021en

MVTec 3D Anomaly Detection Dataset

4,147 high-resolution 3D point cloud scans and RGB images across 10 categories. First comprehensive 3D anomaly detection benchmark.

Metrics:auroc, pixel-auroc, pro
Paper / WebsiteDownload
Current State of the Art

CMDR-IAD

Research

97.3

auroc

MVTec 3D-AD — auroc

6 results · 3 SOTA advances · higher is better

All results
SOTA frontier
85909520222023202420252026aurocBTF (Back to the Feature)M3DMCMDR-IAD

auroc Progress Over Time

Showing 3 breakthroughs from Mar 2022 to Mar 2026

85.488.791.995.198.4Mar 2022Feb 2024Mar 2026aurocDate

Key Milestones

Mar 2022
BTF (Back to the Feature)

Image-level I-AUROC, 2D+3D setting. Reported in CMDR-IAD Table 1.

86.5
Mar 2023
M3DM

Image-level I-AUROC on MVTec 3D-AD. Table 1 in original paper.

94.5
+9.2%
Mar 2026
CMDR-IADCurrent SOTA

Image-level I-AUROC, proposed method. Table 1 in paper.

97.3
+3.0%
Total Improvement
12.5%
Time Span
4y 1m
Breakthroughs
3
Current SOTA
97.3

Top Models Performance Comparison

Top 6 models ranked by auroc

auroc1CMDR-IAD97.3100.0%2MTSJM95.798.4%3CFM95.498.0%43D-ADNAS95.197.7%5M3DM94.597.1%6BTF (Back to the Feature)86.588.9%0%25%50%75%100%% of best
Best Score
97.3
Top Model
CMDR-IAD
Models Compared
6
Score Range
10.8

aurocPrimary

Related Papers3

Other Anomaly Detection Datasets

MVTec 3D-AD Benchmark - Anomaly Detection | CodeSOTA