Segmentation Fault (Core Dumped) about particle-sfm HOT 4 CLOSED

Hi, this is the issue about your GPU driver. It is nothing to do with our repo. Please test the following codes:

import torch
x = torch.rand(5, 3).cuda()
print(x)

See https://stackoverflow.com/questions/54264338/why-does-pytorch-not-find-my-nvdia-drivers-for-cuda-support or https://discuss.pytorch.org/t/found-no-nvidia-driver-on-your-system-but-its-there/35063 to find some possible solutions.

from particle-sfm.

Hi, I made some minor revisions on the code to enforce everything run on CPU, and I met the same issue.
I identified where the segmentation fault comes from, and it arrives at the optimize_location function in the trajectory_optimize.cpp file.
In that function, it is using the ceres solver and does some optimization task, so I don't think the segmentation fault has to do with GPU driver (or maybe I am wrong).

Any thought on this?

from particle-sfm.

@C-H-Chien It is nothing to do with the GPU driver in trajectory optimization. Maybe we can open a new issue to discuss it. BTW, we haven't tested the CPU version yet.

from particle-sfm.

@C-H-Chien Some suggestions:

1. Disable the trajectory optimization to test if all the other part works normally. (--skip_path_consistency)
2. Print the inputs of the trajectory optimization to check if something goes wrong.
3. Check the flow outputs after you modify it to CPU mode.

from particle-sfm.