3d point clouds

[JianGangZH]

Great work, but I have a small question: The pipeline renders the correct image, but the 3D structure is not accurate. Could this be due to the following reason? When using EndoNeRF-Pulling soft tissue data, the method for loading data in GS-Splatting first estimates the camera pose using COLMAP, generating files like camera.bin and images.bin. However, since this is a dynamic dataset and the camera pose does not move, COLMAP may still produce inaccurate poses and initial point clouds. This results in inaccurate initial point clouds being used for GS-Splatting. Even though the deformation network renders the correct image, the point cloud’s shape does not match the actual 3D structure of the tissue.