cuda_path is set but cuda wasn't able to be loaded

[Karl48071]

After doing the install on windows (CUDA, cudnn, onnxruntime-gpu), I get the following error when running deface... CUDA_PATH is set but CUDA wasn't able to be loaded.

I've watched countless videos on setting correct PATHs and so forth and am continuing to have the issue.

[mdraw]

This may be related to https://github.com/microsoft/onnxruntime/issues/13576 If you haven't tried that already I'd suggest installing it in a conda environment with python, cuda and cudnn already installed.

[mdraw]

I've recently tested deface on Windows and found that the new DirectML execution provider also works quite well. It should work on any GPU since it is based on the built-in Direct3D12 API of Windows. I don't know how it compares to CUDA in terms of speed but if you are still having issues with CUDA it may be worth a shot. You can install it with

$ pip install onnx onnxruntime-directml

[HaukeCornell]

I had the same issue (windows). I tried multiple versions, and environment setups. I only got it running by installing a new anaconda environment and using anacondas supplied version of cuDDN and CUDA.

[StealUrKill]

After doing the install on windows (CUDA, cudnn, onnxruntime-gpu), I get the following error when running deface... CUDA_PATH is set but CUDA wasn't able to be loaded.

I've watched countless videos on setting correct PATHs and so forth and am continuing to have the issue.

python -m pip install --upgrade setuptools pip python -m pip install deface pip install nvidia-pyindex pip install tf2onnx pip install onnx onnxruntime-gpu ## if gpu with cuda exist pip install onnxruntime-openvino ## intels acceleration

replace the v11.4 with the verison you have installed....IIRC 11.8 is max that this tool can use at the moment.

python -m pip install "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.4\graphsurgeon\graphsurgeon-0.4.5-py2.py3-none-any.whl" python -m pip install "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.4\uff\uff-0.6.9-py2.py3-none-any.whl" python -m pip install "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.4\onnx_graphsurgeon\onnx_graphsurgeon-0.3.12-py2.py3-none-any.whl" python -m pip install "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.4\python\tensorrt-8.2.3.0-cp39-none-win_amd64.whl"

to use tensorrt

python import tensorrt

[StealUrKill]

you need cuda 11.X - the 8.9.4 works for me so far https://developer.nvidia.com/rdp/cudnn-archive

if you want tensor as will the 8.6.16 GA version works https://developer.nvidia.com/nvidia-tensorrt-8x-download

you will need the latest cuda 11.X toolkit --- its large at 3gb https://developer.nvidia.com/cuda-toolkit-archive

[vertexgamer]

I've recently tested deface on Windows and found that the new DirectML execution provider also works quite well. It should work on any GPU since it is based on the built-in Direct3D12 API of Windows. I don't know how it compares to CUDA in terms of speed but if you are still having issues with CUDA it may be worth a shot. You can install it with

$ pip install onnx onnxruntime-directml

thanks, this worked