I noticed that the function apply_se3_augmentation_with_camera_pose augments the point cloud and action without updating rendering-related parameters (e.g., nerf_target_camera_extrinsic). Could this cause misalignment in GS rendering, or is there an alignment step I may have missed? I’d appreciate any insights you can share.
Additionally, I’m curious why ManiGaussian randomly selects a single view out of the 21 available views for computing the rendering loss instead of utilizing all 21 views. Is there a specific reason for this design choice?
Thank you for this great project, @GuanxingLu!
I noticed that the function
apply_se3_augmentation_with_camera_poseaugments the point cloud and action without updating rendering-related parameters (e.g.,nerf_target_camera_extrinsic). Could this cause misalignment in GS rendering, or is there an alignment step I may have missed? I’d appreciate any insights you can share.Additionally, I’m curious why ManiGaussian randomly selects a single view out of the 21 available views for computing the rendering loss instead of utilizing all 21 views. Is there a specific reason for this design choice?
Thank you!