RADAR produces anomaly maps directly from attention-based diffusion models in a single forward pass, achieving higher F1 scores than reconstruction-based diffusion and statistical baselines on MVTec-AD and 3D-printed material data.
In-situ monitoring of image texturing via random forests and clustering with applications to additive manufacturing
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Single-Step Reconstruction-Free Anomaly Detection and Segmentation via Diffusion Models
RADAR produces anomaly maps directly from attention-based diffusion models in a single forward pass, achieving higher F1 scores than reconstruction-based diffusion and statistical baselines on MVTec-AD and 3D-printed material data.