Troubleshooting

Platform related
Bad dependencies resolutions
Make sure to run Python 3.10 to prevent dependencies conflicts.
No module named 'xxx'
A ModuleNotFoundError indicates that dependencies have not been correctly installed.
pip install -r requirements.txt
Module 'xxx' has no attribute 'xxx'
A AttributeError indicates that dependencies have not been correctly installed.
pip install -r requirements.txt
Solution to most OpenCV errors
In most cases it helps to reinstall the opencv-python dependencies.
pip uninstall opencv-python opencv-python-headless
pip install opencv-python
Solution to most Protobuf errors
In most cases it helps to reinstall all dependencies that provide a model like insightface and opennsfw2.
DNN library is not found
Make sure to have the latest CuDNN runtime installed as third party libraries might be compiled with a higher version.
Illegal instruction
This might be caused by a missing instruction set of the CPU. You can either compile tensorflow from source or downgrade it.
pip install [email protected]
Usage related
NoneType object has no attribute shape
Make sure to remove special characters and unicode in -s/--source-path, -t/--target-path and -o/--output-path arguments.
Cannot allocate memory
Memory exceptions during processing are usually cause when the GPU runs out of VRAM. Try to lower the --execution-thread-count to 1 and increase it slowly.
Leaked semaphore objects
Leaked semaphore during processing are usually cause when the GPU runs out of VRAM. Try to lower the --execution-thread-count to 1 and increase it slowly.
Only CPU being used
Do not install onnxruntime and onnxruntime-xxx at the same time.
pip uninstall onnxruntime onnxruntime-xxx
pip install onnxruntime-xxx
Platform benchmark
Linux
Last modified 1mo ago