search

autonomousvision / occupancy_networks

This repository contains the code for the paper "Occupancy Networks - Learning 3D Reconstruction in Function Space"
https://avg.is.tuebingen.mpg.de/publications/occupancy-networks
MIT License
1.48k stars 291 forks source link

My program is running smoothly, but the generated.off file is empty #129

Closed hahahahahahahap closed 7 months ago

hahahahahahahap commented 8 months ago

run python generate.py configs/demo.yaml It looks like it worked

https://s3.eu-central-1.amazonaws.com/avg-projects/occupancy_networks/models/onet_img2mesh_3-f786b04a.pt => Loading checkpoint from url... Warning: generator does not support pointcloud generation. 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [00:01<00:00, 5.77it/s] Timings [s]: mesh time (encode inputs) time (eval points) time (marching cubes) class name
n/a 0.168913 0.001427 0.034569 0.131687 mean 0.168913 0.001427 0.034569 0.131687

It does generate the.off file, but there's nothing in the.off file.

hahahahahahahap commented 8 months ago

pytorch 1.0.0 cudatoolkit 10.0.130 torchvision 0.2.0 and OS is Ubuntu 22.04.1 Thank you in advance for your help.

Nkcemeka commented 8 months ago

@hahahahahahahap I had the same issue. Have you been able to fix this?

@LMescheder Can you help with this? That will be highly appreciated.

hahahahahahahap commented 8 months ago

@hahahahahahahap I had the same issue. Have you been able to fix this?

@LMescheder Can you help with this? That will be highly appreciated.

You can try running this code with the CPU, which will solve the problem temporarily.

The main issue is probably that my graphics card version does not support the low version of cuda. I have a 3070 graphics card that does not support cuda 10 version.

hahahahahahahap commented 8 months ago

you can change the code with device = torch.device("cpu").

But you know this might not be a good idea. You need to find the right version of cuda for your graphics card as well as pytorch and torchvision... Version of.

Nkcemeka commented 8 months ago

Okay, thanks. I will give that a shot.

I am using a NVIDIA GeForce RTX 4090 with CUDA 12.2. I did have issues and so created a container with CUDA 10. This allowed me get certain things working until I encountered what you experienced.

I will go ahead to use the cpu instead.

Thanks.

Nkcemeka commented 8 months ago

@hahahahahahahap Thanks again. By the way, I just tried it and it worked. I don't know if you have tried training, but if you did, I presume you set the device to cpu. I am not sure I really grab how all these GPU architectures work. Shouldn't recent graphic cards be backward compatible with old CUDA versions? I thought with a separate CUDA toolkit in a Docker container, I could get everything to work without problems. Now, I am lost as to what step to take.

hahahahahahahap commented 8 months ago

@hahahahahahahap Thanks again. By the way, I just tried it and it worked. I don't know if you have tried training, but if you did, I presume you set the device to cpu. I am not sure I really grab how all these GPU architectures work. Shouldn't recent graphic cards be backward compatible with old CUDA versions? I thought with a separate CUDA toolkit in a Docker container, I could get everything to work without problems. Now, I am lost as to what step to take.

My guess is that a higher cuda version (which must be supported by your graphics card's computing power) and a suitable pytroch version will make it work.

I didn't do any further research on this network, and yes, I didn't try to train it, which required plenty of time and hard disk space.

I found a replacement, and I thought this repository might help you with your current research.

Nkcemeka commented 8 months ago

@hahahahahahahap Thanks a lot. I will check it out.