Could you provide a docker please？ about softgroup HOT 13 CLOSED

Cause I meer problems in Installation, I use RTX 3080Ti, it can only support cuda >11. But If I use cuda>11, I will meet problem when I compile the spconv library and compile the external C++ and CUDA ops.

from softgroup.

did you compile current code with pytorch 1.1 and spconv1.0. The PR #16 will provide code on pytorch 1.10 cuda 10.2+ and spconv2x.

from softgroup.

yes，I have compiled the code with torch1.1 and spconv1.0 successfully days ago, but torch 1.1.0 is too old for my GPU(maybe).

from softgroup.

When I run the code to the step "model=model.cuda()", the code is unresponsive for a long time

from softgroup.

Could you try the following snippet to see whether the code hang or not

import spconv
layer = spconv.SubMConv3d(32, 32, 3)
layer = layer.cuda()

from softgroup.

Ok, I will try. Thank you very much!

from softgroup.

If the error still exists, you can wait for the PR release.

from softgroup.

I test the above code with pytorch 1.1 and spconv1.0. and cuda 10.0
"layer = layer.cuda()" is unresponsive, and no error was repotred. It just seems like stuking.
By the way, I want to ask, If the new PR support cuda>11? Cause some people said RTX 3080ti could only work with cuda >11.
We appreciate your work very much, we want to use it on our outdoor scene dataset WHU-MLS dataset and find out if the existing works's having problems on outdoor big scene or not.

from softgroup.

"layer = layer.cuda()" have passed successfully after waiting for like 10 minutes.
Is that normal?

from softgroup.

It should be executed right away. You may try to install with pytorch 1.1 spconv 1.0 cuda 11

from softgroup.

I have try torch1.1.0+cuda11.4before，it also met the same problem with ".cuda()".
I guess maybe torch 1.1 is unmatched with my GPU.
Hope the new PR will solve my problem.

from softgroup.

Hello, could you please update the Training and Inference Testing command line please?
Thank you very much!

from softgroup.

Hi. The instructions have been updated.

from softgroup.