dev(narugo): add manbits training script

import os.path

from ultralytics import YOLO

from ..base import _TRAIN_DIR as _GLOBAL_TRAIN_DIR

_TRAIN_DIR = os.path.join(_GLOBAL_TRAIN_DIR, 'manbits_detect')

def train(train_cfg: str, session_name: str, level: str = 'm',

          max_epochs: int = 300, save_per_epoch: int = 10, **kwargs):

    # Load a pretrained YOLO model (recommended for training)

    _last_pt = os.path.join(_TRAIN_DIR, session_name, 'weights', 'last.pt')

    if os.path.exists(_last_pt):

        model, resume = YOLO(_last_pt), True

    else:

        model, resume = YOLO(f'yolov8{level}.pt'), False

    # Train the model using the 'coco128.yaml' dataset for 3 epochs

    model.train(

        data=train_cfg, epochs=max_epochs,

        name=session_name, project=_TRAIN_DIR,

        save=True, save_period=save_per_epoch, plots=True,

        exist_ok=True, resume=resume,

        **kwargs

    )

                        no_optimize: bool = False):

    if os.path.dirname(onnx_filename):

        os.makedirs(os.path.dirname(onnx_filename), exist_ok=True)

    copy(

        yolo.export(format='onnx', dynamic=True, simplify=not no_optimize, opset=opset_version),

        onnx_filename

    )

    _retval = yolo.export(format='onnx', dynamic=True, simplify=not no_optimize, opset=opset_version)

    _exported_onnx_file = _retval or (os.path.splitext(yolo.ckpt_path)[0] + '.onnx')

    copy(_exported_onnx_file, onnx_filename)