dvlab-research/DSGN

RuntimeError: cuda runtime error (209) : no kernel image is available for execution on the device

tomshalini opened this issue · 5 comments

I followed the same instruction as mentoned in README file but getting below error while running test_net.py

command : python3 tools/test_net.py --loadmodel ./outputs/DSGN_car_pretrained/ -btest 4 -d 2-3

Error: RuntimeError: cuda runtime error (209) : no kernel image is available for execution on the device

Environment Details:
Ubuntu 18.04
torch 1.3.0
torchvision 0.4.1
4GPU GeForce RTX 2080 11GB

CudaError_DSGN

Did you build and compile the costvolume kernel successfully? It seems that it cannot find the compiled kernel.

Did you build and compile the costvolume kernel successfully? It seems that it cannot find the compiled kernel.

As mentioned in README file i ran setup.py file successfully. Do i need to compile costvolume separately? Is the same mentioned in README file? If not how should i compile that?

[1]+ Done cd dsgn/utils/rotate_iou && bash compile.sh

Result of python3 setup.py build develop --user:

running build
running build_py
running build_ext
running develop
running egg_info
writing dsgn.egg-info/PKG-INFO
writing dependency_links to dsgn.egg-info/dependency_links.txt
writing top-level names to dsgn.egg-info/top_level.txt
reading manifest file 'dsgn.egg-info/SOURCES.txt'
writing manifest file 'dsgn.egg-info/SOURCES.txt'
running build_ext
copying build/lib.linux-x86_64-3.7/dsgn/_C.cpython-37m-x86_64-linux-gnu.so -> dsgn
Creating /home/shalini/.local/lib/python3.7/site-packages/dsgn.egg-link (link to .)
dsgn 0.1 is already the active version in easy-install.pth

Installed /home/shalini/Models/DSGN
Processing dependencies for dsgn==0.1
Finished processing dependencies for dsgn==0.1

You do not have to compile the costvolume kernel separately. From your message, I think you should delete the build folder and rebuild it with python3 setup.py develop and see what happens. If it is successful, it will creates a .so file in dsgn/ folder.

You do not have to compile the costvolume kernel separately. From your message, I think you should delete the build folder and rebuild it with python3 setup.py develop and see what happens. If it is successful, it will creates a .so file in dsgn/ folder.

Yes , it is resolved now.
Thank you so much for your time.

That's great. :)