Blendmask test img error

[zhepherd]

[zhepherd]

Traceback (most recent call last):
File "demo/demo.py", line 74, in
demo = VisualizationDemo(cfg)
File "/nfs/project/huangzhipeng/tools/opensorce/AdelaiDet/demo/predictor.py", line 37, in init
self.predictor = DefaultPredictor(cfg)
File "/home/huangzhipeng/miniconda3/envs/blendmask/lib/python3.7/site-packages/detectron2/engine/defaults.py", line 184, in init
self.model = build_model(self.cfg)
File "/home/huangzhipeng/miniconda3/envs/blendmask/lib/python3.7/site-packages/detectron2/modeling/meta_arch/build.py", line 21, in build_model
model = META_ARCH_REGISTRY.get(meta_arch)(cfg)
File "/nfs/project/huangzhipeng/tools/opensorce/AdelaiDet/adet/modeling/blendmask/blendmask.py", line 58, in init
pixel_mean = torch.Tensor(cfg.MODEL.PIXEL_MEAN).to(self.device).view(3, 1, 1)
File "/home/huangzhipeng/miniconda3/envs/blendmask/lib/python3.7/site-packages/torch/cuda/init.py", line 149, in _lazy_init
_check_driver()
File "/home/huangzhipeng/miniconda3/envs/blendmask/lib/python3.7/site-packages/torch/cuda/init.py", line 63, in _check_driver
of the CUDA driver.""".format(str(torch._C._cuda_getDriverVersion())))
AssertionError:
The NVIDIA driver on your system is too old (found version 10010).
Please update your GPU driver by downloading and installing a new
version from the URL: http://www.nvidia.com/Download/index.aspx
Alternatively, go to: https://pytorch.org to install
a PyTorch version that has been compiled with your version
of the CUDA driver.

[zhepherd]

my cuda version is CUDA Version 10.0.130
CUDA Patch Version 10.0.130.1

[stan-haochen]

Hi zhepherd, AdelaiDet supports CUDA 10.0.

Please make sure you have compiled AdelaiDet with the same cuda as your python runtime. This error typically happens when the cuda in your python environment is of a lower version.

Also, you can check official Detectron2 demos. If those work, BlendMask should also work.

[zhepherd]

which python version need？？my python version is 3.7

[stan-haochen]

@zhepherd same as Detectron2, python >= 3.6

[zhepherd]

my python is 3.7 why has cuda error? i try as detectron2 install.md,but also has the error! pls help

[zhepherd]

[stan-haochen]

@zhepherd As I said before, it is probably because you did not compile detectron2/AdelaiDet with the same cuda as your python runtime.

Please make sure you have installed the correct precompiled detectron2 or have compiled with the right nvcc.

I also suggest you go to Detectron2's repo for help.

[zhepherd]

i am think i compile detectron2 with the same cuda as my python runtime

[zhepherd]

[stan-haochen]

So how did you install pytorch?

If with conda, please try

conda list | grep torch

and make sure torch is compiled with the same version of CUDA

[zhepherd]

use pip

[stan-haochen]

It seems to me the torch version you installed is not compiled to support CUDA 10.0.

The NVIDIA driver on your system is too old (found version 10010).
Please update your GPU driver by downloading and installing a new
version from the URL: http://www.nvidia.com/Download/index.aspx
Alternatively, go to: https://pytorch.org to install
a PyTorch version that has been compiled with your version
of the CUDA driver.

[zhepherd]

which torch and torchvision should i install ?

[stan-haochen]

if you insist on using cuda 10.0, try pytorch 1.3 with cuda 10.0. Remember to specify the CUDA version when you install.

[zhepherd]

specify cuda version when install ? how to specify? not the sys default cuda 10.0?

[stan-haochen]

Something like this

pip install torch==1.3.1+cu100 torchvision==0.4.2+cu100 -f https://download.pytorch.org/whl/torch_stable.html

or this

conda install pytorch torchvision cudatoolkit=10.0 -c pytorch

If you have further questions, join our discord channel here

https://discord.gg/afxvbrZ

The issues are for bug reports and feature requests. Not general QAs

[zhepherd]

i look my runtime cuda ,find is not the same as sys cuda

[zhepherd]

i solve this problem after update the nvidia driver,but has anther problem

[zhepherd]

thx very much, i solve this problem after reinstall ,thx very much for help!!! @stan-haochen