lizhe00 / AnimatableGaussians

Code of [CVPR 2024] "Animatable Gaussians: Learning Pose-dependent Gaussian Maps for High-fidelity Human Avatar Modeling"
https://animatable-gaussians.github.io/
Other
813 stars 48 forks source link

Possible code errors in raster settings #9

Closed tb2-sy closed 3 months ago

tb2-sy commented 3 months ago

In line 52 of gaussian_renderer.py, camera_center = torch.linalg.inv(extr)[:3, 3]seems to be changed to camera_center = torch.linalg.inv(extr)[3, :3], although judging from the results It seems to have little impact.

lizhe00 commented 3 months ago

torch.linalg.inv(extr)[3, :3] is zero, because the translation vector is at the upper right corner of extr. You can print both torch.linalg.inv(extr)[:3, 3] and torch.linalg.inv(extr)[3, :3] to check them.

tb2-sy commented 3 months ago

torch.linalg.inv(extr)[3, :3] is zero, because the translation vector is at the upper right corner of extr. You can print both torch.linalg.inv(extr)[:3, 3] and torch.linalg.inv(extr)[3, :3] to check them.

Thanks for the reply. I made a mistake. The code is torch.linalg.inv(extr)[:3, 3] instead of using the transposed world_view_transform to find the inversion.