Description
Point cloud shapes have no natural order, making autoregressive generation hard. CanonicalVAE, from ECCV 2022, maps point clouds into ordered sequences via canonical mapping and generates 3D shapes autoregressively.
The repo includes training and inference code, setup instructions, pretrained checkpoints and samples.
Canonical mapping:Ordered point sequences.
Autoregressive:Shapes step by step.
Checkpoints:Ready inference.
The repo includes training and inference code, setup instructions, pretrained checkpoints and samples.
Features
Canonical mapping:Ordered point sequences.
Autoregressive:Shapes step by step.
Checkpoints:Ready inference.
Tags:3d-modeling