Segmentation Fault Occurs in infer.py When Initializing point_encoder in TGS Class

[GitHofee]

Description

I encountered a segmentation fault (core dumped) error while running infer.py, with no additional error information provided. The error occurs in the __init__ function of the TGS class, when executing the following line:

self.point_encoder = tgs.find(self.cfg.pointcloud_encoder_cls)(self.cfg.pointcloud_encoder)

This line is attempting to initialize self.point_encoder with the configuration provided in self.cfg.pointcloud_encoder, but it leads to a segmentation fault and terminates the program unexpectedly.

Environment

• GPU : NVIDIA GeForce RTX 3090 x 4
• Python : 3.8.19
• PyTorch : 2.1.0+cu118

Thank you for your assistance in resolving this issue.

[zouzx]

I haven't encountered this issue before. Could you locate the specific line of the __init__ function of the LocalPoolPointnet that causes this? This function utilizes the torch_scatter package, have you ensured that this package is installed correctly?