Data preparation on MipNerf360 and NerfSynthetic and stack smashing error

[caoduycp03]

2025-07-18 16:12:07,086 - ERROR: Training failed: File sparse/0/point_cloud.ply not found on /home/ubuntu/duycd/triangle-splatting/data/mip_nerf360/kitchen

Can you show me how to construct a point cloud using the MipNerf360 or NerfSynthetic dataset to achieve a well-reconstructed point cloud as initialization? Thank you for your great work.

[caoduycp03]

2025-07-19 16:14:45,444 - INFO: Fetching intrinsic and extrinsic data from transforms_train.json and transforms_test.json
2025-07-19 16:14:45,473 - INFO: Train set size: 100, Test set size: 200
2025-07-19 16:14:45,479 - INFO: Camera extent: 4.93
2025-07-19 16:14:45,590 - INFO: Train resolution: 800x800, Test resolution: 800x800
2025-07-19 16:14:45,591 - INFO: Using alpha mask: False.
2025-07-19 16:14:46,137 - INFO: Initializing Triangles from point cloud
2025-07-19 16:14:46,137 - INFO: Fetching point cloud data from /home/ubuntu/triangle-splatting/data/blender/ship
/point_cloud.ply.
2025-07-19 16:14:47,563 - INFO: Running inside sampling
2025-07-19 16:14:47,564 - INFO: Sampled 330000 points from 330000 points using direct sampling
2025-07-19 16:14:47,580 - INFO: Running outside sampling
2025-07-19 16:14:47,580 - INFO: Sampled 0 points from 0 points using direct sampling
*** stack smashing detected ***: terminated

I used the output of 3DGS on MipNerf360 as initialization for triangle splatting as in #2 , but it shows errors. Can you help me?

The line scaling = inter_point_distance(points)[..., None] is causing this error, and the issue originates from simple_knn

[Developer-KS]

Can you debug the points tensor and make sure that it is a tensor of shape (N, 3) with no value of nan/inf? (N being the same as the number of points)