Yolo模型Fine-tuning

1.图像采集

 参考《从摄像头采集图像》：https://blog.csdn.net/2301_80049844/article/details/157290607

2.数据标注

参考《数据标注.Labelme》：https://blog.csdn.net/2301_80049844/article/details/157290870

3.格式转换

参考《labelme转yolov8数据格式》：https://blog.csdn.net/2301_80049844/article/details/157291089

4.Fine-tuning

4.1.命令行参数解析

# -*- coding:utf-8 -*-

import argparse

def parse_arguments():
parser = argparse.ArgumentParser()
parser.add_argument(
"–yolo_pt",
required=True,
help="yolo模型文件")

parser.add_argument(
"–coco_yaml",
required=True,
help="coco.yaml文件")

parser.add_argument(
"–train_output",
required=True,
help="训练输目录")

return parser.parse_args()

4.2.Fine-tuning

# -*- coding:utf-8 -*-

import os

import ultralytics

from ultralytics import YOLO

from setting import parse_arguments

os.environ['ULTRALYTICS_OFFLINE'] = 'True'

def train():
args = parse_arguments()

yolo_pt = args.yolo_pt
coco_yaml= args.coco_yaml
train_output = args.train_output

model = YOLO(yolo_pt)
results = model.train(data=coco_yaml,
epochs=200,
batch=4,
device='0',
project=train_output,
name='yolov8_retrain',
exist_ok=False, # True,
amp=False,
)

# 训练完成后，手动指定 val 的保存路径
metrics = model.val(project=train_output, name='yolov8_retrain_val')

if __name__ == "__main__":
train()