Error when testing about snake HOT 14 CLOSED

I deleted and reinstalled the entire project and finally it worked. Maybe I got problem when building.
I confirm that snake can be run on CUDA 0.1 + Pytorch 1.4.
Thanks for your support.

from snake.

What are the versions of your system cuda and pytorch cuda?

from snake.

The version of my system cuda is 10.2; my pytorch cuda is 10.1.

from snake.

Make sure two points:

1. pytorch version that you used to compile the code should be the same as the one you use to run the code.
2. The version of pytorch cuda should be the same as that of the system cuda.

from snake.

But I'm running your project on a server GPU which installed CUDA 10.2 already while the Pytorch official website only provide the introduction for installing CUDA upto 10.1.
Does anyone know how to install Pytorch with CUDA 10.2 from source? Thanks.

from snake.

Are you sure that the pytorch version that you used to compile the code is the same as the one you use to run the code?

from snake.

@pengsida how can I check these versions? Sorry, I'm just a newbie.
Maybe the pytorch version which I used to compile the code is 1.4.0 and cuda 10.1
When I run setup.py in apex, the terminal display something with these lines at the beginning:

torch.__version__  =  1.4.0
setup.py:46: UserWarning: Option --pyprof not specified. Not installing PyProf dependencies!
  warnings.warn("Option --pyprof not specified. Not installing PyProf dependencies!")

Compiling cuda extensions with
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:07:16_PDT_2019
Cuda compilation tools, release 10.1, V10.1.243
from /usr/local/cuda/bin

from snake.

python -c "import torch; print(torch.version.cuda)"
nvcc --version

from snake.

python -c "import torch; print(torch.version.cuda)"
nvcc --version

I ran these two command and recieved 2 responses, respectively:
10.1

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:07:16_PDT_2019
Cuda compilation tools, release 10.1, V10.1.243

But when I ran the testing command, the error still happened. Any idea?

from snake.

How about re-compling this module: python setup.py build_ext --inplace.

from snake.

How about re-compling this module: python setup.py build_ext --inplace.

I re-compiled dcn_v2, extreme_utils, roi_align_layer followed the INSTALL.md but the error still appeared.

from snake.

Did you set export CUDA_HOME="/usr/local/cuda-9.0?
If your cuda is 10.0, it might cause errors.

from snake.

No. Depend on my GPU, I set CUDA_HOME="/usr/local/cuda".

from snake.

You could test pytorch 1.1 and cuda 9.0, as described in INSTALL.md, to see if this problem is caused by the pytorch version and cuda version.

from snake.