search

ShenhanQian / GaussianAvatars

[CVPR 2024 Highlight] The official repo for "GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians"
https://shenhanqian.github.io/gaussian-avatars
Other
613 stars 93 forks source link

dimensions of self.binding #39

Closed xz0305 closed 7 months ago

xz0305 commented 7 months ago

The initial dimension of binding is 10144, which changes to 62451 after the following step. Is this caused by the number of point clouds? self.binding = torch.tensor(binding, dtype=torch.int32, device="cuda").squeeze(-1)

ShenhanQian commented 7 months ago

The first dimension of self.binding should be consistent with the number of points. As the number of points grows during the optimization, self.binding will also grow.

However, this change of size is caused by Adaptive Density Control instead of the line you quoted.