rgbd-kinect-pose
rgbd-kinect-pose copied to clipboard
Using google drive docker image - kinect test, transform engine init failure
Hi @rmbashirov,
Thanks for your support on this repo! It's very interesting project.
I believe my issue may be related to #5.
I have downloaded the docker image from google drive link and I'm attempting to test the kinect viewer. I'm receiving the following error.
I completed the steps indicated by @xgen in #5 to expose all GPU's in the call to ./run_local.sh
.
My GPU is listed when calling nvidia-smi
inside docker, however I notice the CUDA version is 11.6, not 10.2.
Any ideas?
@rmbashirov for posterity, this is related to ampere GPU architecture compatibility with CUDA 10.2. Hopefully I can sort out using this
cheers, David
@rmbashirov for posterity, this is related to ampere GPU architecture compatibility with CUDA 10.2. Hopefully I can sort out using this
cheers, David
Hey did you solve this? It looks like I meet the same issue and I try some methods but none of them works...
@felixshing Alas, I moved on to another project.
@dr00b Thank you for your reply. I also tried it in my 3080 GPU with 11.7 CUDA but I failed for now. I decided to try one more approach. If that still does not work, I will give up.
May I ask did you find some other project related to this (map the keypoints extracted from RGB-D sensors to SMPL-X)?
@felixshing I did not. It was a semester project so I had to cut and run with something different
Oh, thanks for letting me know. I did my last attempt but it still does not work. Thus I just give up. It seems that the real problem comes from the face that the k4a 0.13.0 used in this project does not support 30X GPU well. I try to use the newest k4a package but somehow it must require 0.13.0