ONNX runtime support

[karl-joan]

Made changes to pull request #904 since the author didn't respond.

Changes:

• Separated functions for image preprocessing and post-processing.
• You can supply options to onnxruntime.
• Fixed bugs in image post-processing, where iou_threshold was missing and bounding boxes were sent to non_max_supression in wrong format.
• Fixed bug where sessions options weren't passed to onnxruntime.
• Added some test.
• Renamed onnx.py to onnx_model.py to avoid confusion with onnx library.

The code was designed with the intention of being used for YOLOv8 ONNX model. Perhaps some changes should be made to make it more suitable for other architectures.

[julled]

Thanks for continuing this work! I thinks its very important and i am looking forward to test this if you think its ready.

[karl-joan]

@julled You can go ahead and test it

[LongLe18]

@karl-joan I tested your contribution. It's appear error like image below. And image which I used to test (https://ultralytics.com/images/bus.jpg)

[karl-joan]

@LongLe18 The test is meant to be run with image specified in the test file. If you change the image, then it won't work since the image you provided does not contain any cars.

If you just want to try the model, then use one of the built in functions such as get_sliced_prediction or get_prediction.

[LongLe18]

@karl-joan ok tks sir. I understood, you have not handled except yet. I will try again

[fcakyon]

Thanks a lot for this amazing contribution @karl-joan !

If you could add a demo notebook as other model implementations, we would gladly merge the PR 👍

[fcakyon]

code styling errors need to be fixed, a demo notebook need to be added

[fcakyon]

@karl-joan can you please fix the styling? https://github.com/obss/sahi#contributing

[karl-joan]

@fcakyon Yes, I'll try to get that and the notebook tomorrow done.

[karl-joan]

@fcakyon Let me know if you need anything else.

[devrimcavusoglu]

@karl-joan @fcakyon what is the status here ? I can take the lead, but I assume code formatting/linting has been addressed. As far as I reviewed the files, there are no additional requirements.

[karl-joan]

@devrimcavusoglu It's currently waiting behind @fcakyon

[devrimcavusoglu]

LGTM! Thank you for the contribution @karl-joan