环境准备

这里我的环境是Windows 环境

YOLO v8

下载链接：https://github.com/ultralytics/ultralytics

官方教程链接：https://docs.ultralytics.com/quickstart/

预训练模型：https://github.com/ultralytics/assets/releases

创建自己的数据集

1.首先准备了VOC 格式的数据集

推荐使用labelImg 得到 Annotations
数据文件结构
-data
—images 所有图片（.jpg .png 等）
—Annotations VOC格式标注（.xml）
—ImageSets 目前为空
—labels 目前为空

2.然后确定用于训练、测试的数据

import os
from random import sample
from pathlib import Path

trainval_percent = 0
train_percent = 1
xml_file_path = Path('data/Annotations')
txt_save_path = Path('data/ImageSets')
total_xml = os.listdir(xml_file_path)

num = len(total_xml)
num_list = list(range(num))
tv = int(num * trainval_percent)
tr = int(tv * train_percent)
trainval = sample(num_list, tv)
train = sample(trainval, tr)

with open(txt_save_path / 'trainval.txt', 'w') as ftrainval, \
        open(txt_save_path / 'test.txt', 'w') as ftest, \
        open(txt_save_path / 'train.txt', 'w') as ftrain, \
        open(txt_save_path / 'val.txt', 'w') as fval:

    for i in num_list:
        name = total_xml[i][:-4] + '\n'
        if i in trainval:
            ftrainval.write(name)
            if i in train:
                ftest.write(name)
            else:
                fval.write(name)
        else:
            ftrain.write(name)

3.将VOC格式标注转为YOLO 标注

import xml.etree.ElementTree as ET
import os

sets = ['train']
classes = ["WCF"]

def convert(size, box):
    dw = 1. / size[0]
    dh = 1. / size[1]
    x = (box[0] + box[1]) / 2.0
    y = (box[2] + box[3]) / 2.0
    w = box[1] - box[0]
    h = box[3] - box[2]
    x *= dw
    w *= dw
    y *= dh
    h *= dh
    return (x, y, w, h)

def convert_annotation(image_id):
    in_file = open(f'data/Annotations/{image_id}.xml')
    out_file = open(f'data/labels/{image_id}.txt', 'w')
    tree = ET.parse(in_file)
    root = tree.getroot()
    size = root.find('size')
    w = int(size.find('width').text)
    h = int(size.find('height').text)

    for obj in root.iter('object'):
        difficult = obj.find('difficult').text
        cls = obj.find('name').text
        if cls not in classes or int(difficult) == 1:
            continue
        cls_id = classes.index(cls)
        xmlbox = obj.find('bndbox')
        b = (
            float(xmlbox.find('xmin').text),
            float(xmlbox.find('xmax').text),
            float(xmlbox.find('ymin').text),
            float(xmlbox.find('ymax').text)
        )
        bb = convert((w, h), b)
        out_file.write(f'{cls_id} {" ".join(str(a) for a in bb)}\n')
    in_file.close()
    out_file.close()

wd = os.getcwd()
print(wd)
for image_set in sets:
    os.makedirs('data/labels/', exist_ok=True)
    with open(f'data/ImageSets/{image_set}.txt') as f:
        image_ids = f.read().strip().split()
    with open(f'data/{image_set}.txt', 'w') as list_file:
        for image_id in image_ids:
            list_file.write(f'data/images/{image_id}.jpg\n')
            convert_annotation(image_id)

4.配置数据文件 yaml

自己创建 data.yaml

# train val test 集合
train: .../data/train.txt
val: .../data/train.txt
test: .../data/train.txt

# 类别标签
names:
  0: tree

# nc: 类别数
nc: 1

配置 YOLO v8

需要找到yolo v8 的yaml 文件
通常地址为：ultralytics\models\v8
修改类别数
也可改动里面其他的信息

安装和训练

安装依赖包

pip install ultralytics

训练

Win 上workers 尽量改为1

if __name__ == '__main__':
    from ultralytics import YOLO

    # Load a model

    model = YOLO('.../ultralytics/models/v8/yolov8m.yaml').load('models/yolov8m.pt')  # build from YAML and transfer weights

    # Train the model
    model.train(data='data/data.yaml', batch=3, epochs=300, imgsz=640,workers=1)