Can we use this approach to embed point clouds？

[misakamage]

Thank you for open-sourcing this excellent work! I have recently been thinking about schemes for embedding point clouds into feedforward neural networks while preserving complete information. I believe point clouds are, in a sense, a degenerate form of 3D Gaussian Splatting (3DGS). Do you think it would be feasible to apply this work to point cloud embedding?

[yuelinxin]

For point clouds, it probably make less sense to embed each point individually. I'm sure there's some suitable way to embed a local group of points as an embedding, that suits your need.