Closed ZisongXu closed 6 months ago
@ZisongXu, I'm not familiar with singularity. It might worth checking here and revise the cuda version that matches your GPU. Then rebuild the docker image.
Also, I'd recommend to take a look at our recent work FoundationPose, which does not need per-object training and can be immediately applied to novel objects.
@wenbowen123 Ok thanks a lot!!! Can I ask when will you upload the code of FoundationPose, I realized that I can see the paper on arXiv.
@wenbowen123 Oh sorry, by the way do I also need to change the version of Ubuntu from 16 to 20?
Our target release is in March/April.
Re Ubuntu version, I haven't tried that. If you run into dependency issues that expect certain OS version, might worth changing that too.
Thanks a lot!
@ZisongXu Our FoundationPose code has released.
Hi Dear Bowen:
I am coming from your another repository "BundleSDF", and you recommend me to use "se3-tracknet".
I am trying re-run your code but I got this issue:
RuntimeError: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location=torch.device('cpu') to map your storages to the CPU.
Because our lab uses the singularity container, I would like to reproduce your project in singularity first. When I built the singularity container, I changed the image to the docker image you provided and it seemed that the build was successful. But when I run your example code it had the above problem.
I am thinking that my GPU is 4090, could it be that the version of cuda and cudnn is too low, causing the GPU to be unable to be used? Because after I created the container, I found that it was ubuntu16 and the cuda version was 10.1.
If your code supports GPU4090, maybe I should use docker containers. But I just wanted to try singularity first.
Best Regards Zisong Xu