Added yolov8engine.py for TensorRT-quantized YOLOv8 models

[alsozatch]

Model file to allow for SAHI inference on yolov8 TensorRT quantized .engine models. Used exactly the same as the base Yolov8 model, except

1. Requires a config_path variable to a .json file with below structure.
2. Can't use the config file fields as extra fields in **kwargs like the base yolov8.py uses config_path.

Here's a function to grab the config file information from the .pt file that you quantized from initially.
Still need to manually put in imgsz and half flag equivalent to the imgsz and half parameters you use to quantize it.

def record_metadata_json(self, json_path, imgsz, half):
        task = self.model.task
        names = self.model.names
        names_list = list(names.values())

        metadata = {
            "task": task,
            "names": names_list,
            "imgsz": imgsz,
            "half": half,
        }
        
        with open(json_path, 'w') as json_file:
            json.dump(metadata, json_file, indent=4)

[fcakyon]

@alsozatch, thanks for your contribution!

Can you please check the errors in the tests and try to fix them?

[alsozatch]

@alsozatch, thanks for your contribution!

Can you please check the errors in the tests and try to fix them?

It looks like you are using 'isort' and 'black' to enforce some order for imports. I've not used these packages so I'm not sure what I'm looking at. Is the required import format specified somewhere?

[fcakyon]

@alsozatch you can check this section, formatting is very easy and automated by a script 👍🏻 https://github.com/obss/sahi?tab=readme-ov-file#contributing

[lucasbarrez]

Hey @alsozatch, @fcakyon,

Will the integration and maintenance of the use of the engine format be available? Or should we implement it on our side?

Thx.