GPU version of the program cannot load the model

[zdpaa]

I have downloaded the dependencies and the corresponding CUDA for the GPU version, but after running it, the automatic annotation model does not load

[vietanhdev]

Please specify:

• Your OS
• Your PC hardware (CPU/GPU)
• The steps you used to install AnyLabeling
• The error message

Please check the instructions to use GPU here: https://anylabeling.com/docs/gpu.

[zdpaa]

@zdpaa In the release note of ONNX Runtime 1.14.1, they support CUDA 12. Which version are you using on your machine?

cuda is 11.6 and cudnn is 8.5

[zdpaa]

@zdpaa In the release note of ONNX Runtime 1.14.1, they support CUDA 12. Which version are you using on your machine?
ONNX Runtime 1.14.1 only support CUDA 12.？

[zdpaa]

ntime 1.14.1 only support CUDA 12.？

[vietanhdev]

@zdpaa Could you help me to try ONNXRuntime 1.14.0?

pip install onnxruntime-gpu==1.14.0

[zdpaa]

@zdpaa Could you help me to try ONNXRuntime 1.14.0?

pip install onnxruntime-gpu==1.14.0

[zdpaa]

I tried 1.14.1 and 1.14.0 they are fine, but maybe my computer graphics card configuration is not enough, the model can be loaded, but after loading to mark will keep spinning and then automatically exit

My gpu is 3060, 6g video memory
The cpu version can instead be used normally, although sometimes there will be an automatic exit

[vietanhdev]

@zdpaa My SAM B quantized model takes 7.2GB of GPU memory when running on RTX 2070. Your GPU memory may not be enough.
However, the issue may be because the model download was not successful. I will release v0.2.23 to handle model downloading better. In the meantime, you may want to download the models from here, extract and load them into your program using instructions from this page.
Thank you for your feedback!