A ViT-based masked reconstruction model trained on normal data detects and localizes anomalies under arbitrary viewpoints from as few as two reference images, without 3D reconstruction.
ICLR (2021)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
PADFormer: Pose-agnostic Anomaly Detection from Sparse View Images
A ViT-based masked reconstruction model trained on normal data detects and localizes anomalies under arbitrary viewpoints from as few as two reference images, without 3D reconstruction.