nvrtc: error: invalid value for --gpu-architecture (-arch)

[520jz]

Thank you very much for your work, I got an error when I ran the code below as per your tutorial:

my gpu is RTX4090. my cuda is 11.6 and pytorch is 1.12.1

[marcocannici]

Hi @520jz. The issue you are reporting seems related to this pytorch issue. RTX4090 requires compute capabilities >=sm_8.9, which should be supported at CUDA 11.8 or above. You should be able to fix the issue by installing pytorch >= 2.0.0. I tested this code on pytorch 1.13.1 only, so some minor modifications might be needed to make it run on a newer version.

Please let me know if a newer pytorch version solves the problem.

[520jz]

Hi @520jz. The issue you are reporting seems related to this pytorch issue. RTX4090 requires compute capabilities >=sm_8.9, which should be supported at CUDA 11.8 or above. You should be able to fix the issue by installing pytorch >= 2.0.0. I tested this code on pytorch 1.13.1 only, so some minor modifications might be needed to make it run on a newer version.

Please let me know if a newer pytorch version solves the problem.

Thank you very much for your answer, I changed cuda and pytorch to 11.8 and 2.0.0 respectively as you suggested, and now it works successfully, but what does this mean? I chose (3), does this affect the effect of the experiment?

[520jz]

Hi @520jz. The issue you are reporting seems related to this pytorch issue. RTX4090 requires compute capabilities >=sm_8.9, which should be supported at CUDA 11.8 or above. You should be able to fix the issue by installing pytorch >= 2.0.0. I tested this code on pytorch 1.13.1 only, so some minor modifications might be needed to make it run on a newer version.

Please let me know if a newer pytorch version solves the problem.

oh！I'd like to ask one more question, is there a cloud file in the dataset you provided? that is, the *.ply file, the ply file generated by colmap:

[520jz]

Hi @520jz. The issue you are reporting seems related to this pytorch issue. RTX4090 requires compute capabilities >=sm_8.9, which should be supported at CUDA 11.8 or above. You should be able to fix the issue by installing pytorch >= 2.0.0. I tested this code on pytorch 1.13.1 only, so some minor modifications might be needed to make it run on a newer version. Please let me know if a newer pytorch version solves the problem.

oh！I'd like to ask one more question, is there a cloud file in the dataset you provided? that is, the *.ply file, the ply file generated by colmap:

I remember that poses_bounds.npy file was converted from these files generated by colmap, do you have these files generated by colmap?

[marcocannici]

Regarding W&B, it is an online experiments logging platform. You can check their website here. Option (3) is okay, but you will not be able to log the training statistics and results. If you want to use an alternative method, Tensorboard is also available in the code, which you can activate by providing the --no_wandb --use_tensorboard arguments.

The experiments reported in the main paper use a linear slider to collect poses and thus did not require using COLMAP. The supplementary material provides additional results obtained using COLMAP for camera pose estimation. In this case, we obtained the poses_bounds.npy file as you described. However, we did not save the COLMAP sparse reconstruction as we did not need it for our method.

[520jz]

Regarding W&B, it is an online experiments logging platform. You can check their website here. Option (3) is okay, but you will not be able to log the training statistics and results. If you want to use an alternative method, Tensorboard is also available in the code, which you can activate by providing the --no_wandb --use_tensorboard arguments.

The experiments reported in the main paper use a linear slider to collect poses and thus did not require using COLMAP. The supplementary material provides additional results obtained using COLMAP for camera pose estimation. In this case, we obtained the poses_bounds.npy file as you described. However, we did not save the COLMAP sparse reconstruction as we did not need it for our method.

I understand. Thank you very much for your answer