yolov8-pose，labelme关键点标注及格式转换

目录

一、代码、权重下载

二、标注规则

三、标注要求及步骤

四、json标签样式

五、json转txt格式

---

一、代码、权重下载

YOLOv8代码：ultralytics/ultralytics首页 - GitCode

环境安装

pip install ultralytics

 

关键点训练权重下载：

https://github.com/ultralytics/assets/releases/download/v0.0.0/yolov8n-pose.pt

https://github.com/ultralytics/assets/releases/download/v0.0.0/yolov8s-pose.pt

https://github.com/ultralytics/assets/releases/download/v0.0.0/yolov8m-pose.pt

https://github.com/ultralytics/assets/releases/download/v0.0.0/yolov8l-pose.pt

https://github.com/ultralytics/assets/releases/download/v0.0.0/yolov8x-pose.pt

 https://github.com/ultralytics/assets/releases/download/v0.0.0/yolov8x-pose-p6.pt

 

二、标注规则

标签组成：类型，1个数据；目标框坐标，4个数据；关键点，n*3个数据（n为关键点个数，'3'为坐标及是否可见，点标签：0代表此点不在图像上，点标签：1   代表此点在图像上且在未遮挡处，点标签：2   代表此点在图像上但在遮挡处）

0 0.07390752032520326 0.5280939476061427 0.1384654471544715 0.22244805781391144 0.14059959349593495 0.5489837398373983 0 0.04024390243902439 0.5354336043360434 2 0.006580284552845528 0.6133468834688347 1 0.11709857723577236 0.6122177055103883 1
0 0.11709857723577234 0.30451671183378504 0.0431910569105691 0.09146341463414637 0.10312499999999998 0.3163730803974706 2 0.10566565040650407 0.32540650406504057 1 0.13678861788617885 0.3220189701897019 0 0.13678861788617885 0.3141147244805781 0
0 0.7535315040650405 0.857813911472448 0.49034552845528445 0.28116531165311665 0.8964430894308942 0.99726738934056 0 0.9967987804878048 0.9679087624209575 0 0.6404725609756097 0.8753161698283649 2 0.5108993902439024 0.9679087624209575 1
0 0.982981220657277 0.212245696400626 0.02787558685446001 0.06585811163275952 0.9807805164319249 0.220396452790819 2 0.9697769953051644 0.220396452790819 0 0.9705105633802817 0.22691705790297342 0 0.9936179577464789 0.22756911841418886 1

三、标注要求及步骤

（1）标注工具：labelme

（2）按顺序先标注车辆长方形包围拉框，再标记车辆ABCD四个顶点（备注 ：若拉框顺序与标点顺序反了，请重新标注）

（3）四个车底盘顶点顺序

    逆时针：A点、B点、C点、D点

（4）顶点标签

    0表示没有显露出不可见，1表示可见，2表示被遮挡不可见

四、json标签样式

{
  "version": "5.0.1",
  "flags": {},
  "shapes": [
    {
      "label": "car",
      "points": [
        [
          8.97560975609756,
          450.2195121951219
        ],
        [
          274.8292682926829,
          690.4634146341463
        ]
      ],
      "group_id": null,
      "shape_type": "rectangle",
      "flags": {}
    },
    {
      "label": "0",
      "points": [
        [
          269.9512195121951,
          592.9024390243902
        ]
      ],
      "group_id": null,
      "shape_type": "point",
      "flags": {}
    },
    {
      "label": "2",
      "points": [
        [
          77.26829268292683,
          578.2682926829268
        ]
      ],
      "group_id": null,
      "shape_type": "point",
      "flags": {}
    },
    {
      "label": "1",
      "points": [
        [
          12.634146341463413,
          662.4146341463414
        ]
      ],
      "group_id": null,
      "shape_type": "point",
      "flags": {}
    },
    {
      "label": "1",
      "points": [
        [
          224.8292682926829,
          661.1951219512194
        ]
      ],
      "group_id": null,
      "shape_type": "point",
      "flags": {}
    }
  ],
  "imagePath": "1127_52.jpg",
  "imageData":
  "imageHeight": 1080,
  "imageWidth": 1920
}

五、json转txt格式

     要求：标注步骤必须按照先拉框，然后按逆时针标注关键点生成json文件

  转换脚本

import json
import os
import argparse
import cv2
from tqdm import tqdm
import numpy as np

#标签类别字典
  # 0: car
dictlist=['car']

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


 
def convert_label_json(json_dir, img_path, save_dir):
    json_paths = os.listdir(json_dir) 
    for json_path in tqdm(json_paths):
        path = os.path.join(json_dir,json_path)
        with open(path,'r',encoding='UTF-8') as load_f:
            json_dict = json.load(load_f)
        img_w =  json_dict['imageWidth']  # 图片的高
        img_h =  json_dict['imageHeight']  # 图片的宽
        txt_path = os.path.join(save_dir, json_path.replace('json', 'txt'))
        with open(txt_path, 'w') as txt_file:
            points_nor_list = []
            fournum=1
            for shape_dict in json_dict["shapes"]:
                if shape_dict["shape_type"]=="rectangle":

                    x1 = float(shape_dict['points'][0][0])
                    y1 = float(shape_dict['points'][0][1])
                    x2 = float(shape_dict['points'][1][0])
                    y2 = float(shape_dict['points'][1][1])
                    bb = (x1, y1, x2, y2)
                    bbox = convert((img_w, img_h), bb)
                    label1=shape_dict["label"]
                    no_0=dictlist.index(label1)
                    line=np.append(str(no_0),bbox)
                if shape_dict["shape_type"]=="point":
                    x0=float(shape_dict['points'][0][0])/img_w
                    y0=float(shape_dict['points'][0][1])/img_h
                    pointlist=[x0,y0,shape_dict["label"]]
                    line=np.append(line,pointlist)
                if fournum%5==0:
                    stringline=' '.join(line)
                    stringline=stringline+'\n'
                    txt_file.writelines(stringline)                    
                fournum+=1

if __name__ == "__main__":
    parser = argparse.ArgumentParser(description='json convert to txt params')
    parser.add_argument('--json-dir', type=str, default=r"C:xxxxxxxxxxxxx", help='json path dir')#标签路径
    parser.add_argument('--img-path', type=str, default=r"C:xxxxxxxxxxxxx", help='json path dir')#图片路径
    parser.add_argument('--save-dir', type=str, default=r"C:xxxxxxxxxxxxx", help='txt save dir')#txt保存路径
    args = parser.parse_args()
    json_dir = args.json_dir
    img_path = args.img_path
    save_dir = args.save_dir
    convert_label_json(json_dir, img_path, save_dir)