nanodet model does not has CPU support for inference

[pyther-hub]

For GSoC 2023, I have chosen the Nanodet model, which currently supports predictions only for GPU devices and not for CPU. I have identified an issue related to this limitation, which is indicated in the code through the use of ".cuda()".

[blaz-r]

I'm not entirely sure, but I think it should also work on the CPU. I think that the issue you linked only specifies that training can't be done on the CPU, but in the case of the notebook you want inference, so I think it should work. The repository also includes OpenVINO section which also has the option to export to ONNX, so maybe look into that.
The issue also says something about removing torch.cuda.synchronize() sections in order for it to work on the CPU as well.
Hope that helps :)

[pyther-hub]

but if I remove that section then I have to change the source code also

[blaz-r]

In some cases you can change only this particular function by overriding it on already instantiated model.
Did you also check OpenVINO section for conversion procedure?

[pyther-hub]

Yes sir I saw that section I am able to convert the model. for the time being I have edited the source code so that it works for cpu

[pyther-hub]

to be precise if we look here there we have cuda.sun I need to remove these lines so this works properly