ESCAPE: Equivariant Shape Completion via Anchor Point Encoding Shape completion, a crucial task in 3D computer vision, involves predicting and filling the missing regions of scanned or partially observed objects. Current methods expect known pose or canonical coo...

Our work introduces a transformer-based encoding with anchor points, enabling rotation-equivariant shape completion. We represent shapes using distances to anchors, ensuring robustness across varying object orientations.

arxiv.org/abs/2412.00952

#ShapeCompletion #Transformers #Equivariance #cvpr