Segmentation fault (core dumped) when training

zju3dv / PGSR

[TVCG2024] PGSR: Planar-based Gaussian Splatting for Efficient and High-Fidelity Surface Reconstruction

https://zju3dv.github.io/pgsr/

Other

508 stars 33 forks source link

Closed guishe666 closed 3 months ago

guishe666 commented 4 months ago

Hi, Thanks for the awesome work! I met a problem when training the model.

It will report "Segmentation fault (core dumped)" when calculating lncc in file loss_utils.py. It occurs when using the function F.conv2d.

I found it probably because the version of PyTorch. Related: Segmentation Fault in PyTorch 2.2 with F.conv3d function I solved by reducing the version of PyTorch to 2.0.0.

my environment info: OS: Ubuntu 22.04 Device: Nvidia RTX 4090 Cuda version: 11.8

danpeng2 commented 4 months ago

Thank you for your feedback. We have used the PyTorch 2.3.1 version and have not encountered this error.

David-Zhao-1997 commented 3 months ago

Encountered the same problem with PyTorch 2.2.1+cu118, upgrading PyTorch 2.4.0 solved the problem.