【保姆级教程】YOLOv8半自动化数据标注

一、YOLOV8环境准备

1.1 下载安装最新的YOLOv8代码

 仓库地址： https://github.com/ultralytics/ultralytics

1.2 配置环境

  pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple

1.3 安装labelme标注工具

 pip install labelme

二、半自动标注

2.1 下载预训练权重yolov8n.pt

仓库：https://github.com/ultralytics/ultralytics
在YOLOv8 github上下载预训练权重：yolov8n.pt，ultralytics\ultralytics\路径下,新建weights文件夹，预训练权重放入其中。

2.2 下载预训练权重ViT-B SAM model

仓库：https://github.com/facebookresearch/segment-anything?tab=readme-ov-file
在SAM github上下载预训练权重：ViT-B SAM model，将下载的权重重命名为sam_b.pt(包括文件后缀)，放入ultralytics\ultralytics\weights文件夹，预训练权重放入其中。

2.3 自动标注

执行下面代码：（建议使用绝对路径）

from ultralytics.data.annotator import auto_annotate

auto_annotate(data='D:/study/cnn/yolo/ultralytics/ultralytics/assets', det_model='D:/study/cnn/yolo/ultralytics/weights/yolov8n.pt', sam_model='D:/study/cnn/yolo/ultralytics/weights/sam_b.pt')

查看ultralytics\assets_auto_annotate_labels路径下，生成的txt格式标签

三、标签文件可视化

3.1 txt标签转json标签

执行下面代码：（建议使用绝对路径）

import torch
import numpy as np
import base64, os
from PIL import Image
import io
import json

def xyn2xyxy(x):
    # Convert nx4 boxes from [x, y, w, h] to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right
    x = [ar.reshape(-1, 2) for ar in x]

    return x

def txt2points(txtfile_path):
    cls, xywh_list = [], []
    with open(txtfile_path, "r") as f:
        for line in f.readlines():
            line = line.strip('\n').split(' ')  # 去掉列表中每一个元素的换行符
            cls.append(line[0])
            xywh_list.append(line[1:])

    return cls, xywh_list


def savejson(points_list, clses_list, img_tmp, filename, save_dir, w, h):
    cur_json_dict = {
        "version": "5.1.1",
        "flags": {},
        "shapes": [
        ],
    }

    listbigoption = []
    for cls, points in zip(clses_list, points_list):
        points = np.array(points, dtype=float).reshape(-1, 2)
        points[:, 0] = points[:, 0] * w
        points[:, 1] = points[:, 1] * h

        cur_json_dict['shapes'].append(
            {"label": cls, "points": points.tolist(), "group_id": None,
             "shape_type": "polygon", "flags": {}})

    # A['imageData'] = base64encode_img(img_tmp)

    cur_json_dict["imageData"] = None
    cur_json_dict["imageHeight"] = h
    cur_json_dict["imageWidth"] = w
    
    # 图像文件的相对路径
    cur_json_dict['imagePath'] = "..//assets//" + filename

    suffix = os.path.splitext(filename)[-1]
    with open(save_dir + "/" + filename.replace(suffix, ".json"), 'w', encoding='utf-8') as f:
        f.write(json.dumps(cur_json_dict, indent=4, separators=(',', ':')))
        # json.dump(A, f, indent=2, ensure_ascii=False)


def txt2json_main():
    img_dir = r'D:/study/cnn/yolo/ultralytics/ultralytics/assets'
    txt_dir = r'D:/study/cnn/yolo/ultralytics/ultralytics/assets_auto_annotate_labels'
    save_dir = r'D:/study/cnn/yolo/ultralytics/ultralytics/assets'
    

    for imgfile in os.listdir(img_dir):
        print(imgfile)
        name, suffix = os.path.splitext(imgfile)
        txtfile = imgfile.replace(suffix, '.txt')
        txt_path = os.path.join(txt_dir, txtfile)
        if not os.path.isfile(txt_path):
            continue
        img_path = os.path.join(img_dir, imgfile)
        img = Image.open(img_path)
        w, h = img.size
        cls, xyxy = txt2points(txt_path)
        # print(cls)
        # print(xyxy)
        # print()
        savejson(xyxy, cls, img, imgfile, save_dir, w, h)
if __name__ == '__main__':
    txt2json_main()

查看生成结果

3.2 使用labelme工具进行可视化