search

mhamilton723 / STEGO

Unsupervised Semantic Segmentation by Distilling Feature Correspondences
MIT License
712 stars 143 forks source link

Problem with environment #43

Open FlyingYanglu opened 2 years ago

FlyingYanglu commented 2 years ago

I'm trying to install this environment on a RTX3070 GPU, but after installation, when I run precompute_knn.py, it always raise this error:

NVIDIA GeForce RTX 3070 with CUDA capability sm_86 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_70. If you want to use the NVIDIA GeForce RTX 3070 GPU with PyTorch.

I've tried to install cudatoolkit 11.7 and 11.3, but I got same error on both of them. I wonder what GPU the script was originally used on and is there a solution to my problem? Thanks in advance.

codingbutstillalive commented 1 year ago

I am having a similar, but unrelated problem.

On my 2019 Macbook with Intel processor, I am getting the following error message when trying to set up the environment:

ResolvePackageNotFound: 
  - cudatoolkit=11.0
  - cupy
  - nvidia-apex==0.1.0

Shall I just skip those? Are there alternatives. As far as I know, there is no CUDA support for the AMD graphics cards.