What the lib "pointops" for about segmentanything3d HOT 2 OPEN

Sorry for not understanding you well. There are some pointclouds operations implemented on cuda in the pointops. What do you mean source?

from segmentanything3d.

Sorry for not expressing the question well. What I mean is, is the pointclouds operations implemented by 'pointops' defined by individuals (i.e. author yourself), or is is some kind of library that was predefined such that we only need to call them? If it is a library stuff, where is it originated?

By the way, I follow the guidelines and run the code, but I dont know why the output pointcloud by SAM3D is somehow lack of groups(or labels? colors?). The final visualization of scene0000_00 is shown below. The running feedback of the terminal is normal (showing 0.jpg to 5560.jpg and merging pointcloud by halfing the number until it goes to 1).

Do you happen to know where could be wrong in the process? Thanks a lot.

And the visualization code I use is:

from util import *
import torch

pcd_filepath = 'F:/sam3d/processed_pcd/train/scene0000_00.pth'
pcd_seg_filepath = 'F:/sam3d/results/pcd/scene0000_00.pth'

pcd_data = torch.load(pcd_filepath)
seg_data = torch.load(pcd_seg_filepath)

save_path = 'F:/sam3d/results/pcd/scene0000_00.ply'
visualize_partition(pcd_data['coord'], dict(group=seg_data)['group'], save_path)

from segmentanything3d.