You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When generating the Gaussian in GSplat, I observed a misalignment with the initial point cloud. The misalignment is noticeable in terms of axes, position, and scale. It does not appear to be a simple axis inversion but rather involves a seemingly random rotation.
Steps to Reproduce:
Load an initial point cloud from colmap (It has proper alignment parallel to the axis since it comes from synthetic data)
Generate the Gaussian it doesn't matter if you use Default or MCMC strategies.
Compare the Gaussian's orientation, position, and scale with the original point cloud.
Expected Outcome: The Gaussian should align in position, orientation, and scale with the initial point cloud.
Actual Outcome: The Gaussian exhibits an undesired rotation and misalignment in axes, position, and scale relative to the point cloud. The result seems to be coherent between trainings and strategies.
The text was updated successfully, but these errors were encountered:
Are you comparing gaussians with the colmap's pointcloud after the optimization or before? Because after the optimization process it is expected to have different number of gaussians, diffeerent positions, colors, opacities, etc. since it is what actually was optimized
However, centers of the gaussians before optimization (right after create_splats_with_optimizers function) should align with the ones in the pointcloud. Is that the case?
Or do you mean a global rotation/scale of the whole pointcloud?
When generating the Gaussian in GSplat, I observed a misalignment with the initial point cloud. The misalignment is noticeable in terms of axes, position, and scale. It does not appear to be a simple axis inversion but rather involves a seemingly random rotation.
Steps to Reproduce:
Load an initial point cloud from colmap (It has proper alignment parallel to the axis since it comes from synthetic data)
Actual Outcome: The Gaussian exhibits an undesired rotation and misalignment in axes, position, and scale relative to the point cloud. The result seems to be coherent between trainings and strategies.
The text was updated successfully, but these errors were encountered: