retinaface数据集制作

*项目需要，第一次接触，简单做个记录，无脑代码

• 用labelme做的标记，按照顺序：五个点+一个bbox
• 生成.json的文件
• 下面代码将json标记的文件夹转为Retinaface训练集的label.txt

import os
import json

data_dir = 'train标注\\'
all_json = os.listdir(data_dir)

with open("train.txt","w") as f:
     for j_name in all_json:
          f.write(j_name + '\n')
          j = open(data_dir + j_name, encoding='utf-8'
         
          info = json.load(j)
       
          x1 = info['shapes'][5]['points'][0][0]
          y1 = info['shapes'][5]['points'][0][1]
          x2 = info['shapes'][5]['points'][1][0]
          y2 = info['shapes'][5]['points'][1][1]
          w = str(round(x2 - x1, 2))
          h = str(round(y2 - y1, 2))
          x1 = str(round(x1, 2))
          y1 = str(round(y1, 2))
          d1x = str(round(info['shapes'][0]['points'][0][0], 2))
          d1y = str(round(info['shapes'][0]['points'][0][1], 2))
          d2x = str(round(info['shapes'][1]['points'][0][0], 2))
          d2y = str(round(info['shapes'][1]['points'][0][1], 2))
          d3x = str(round(info['shapes'][2]['points'][0][0], 2))
          d3y = str(round(info['shapes'][2]['points'][0][1], 2))
          d4x = str(round(info['shapes'][3]['points'][0][0], 2))
          d4y = str(round(info['shapes'][3]['points'][0][1], 2))
          d5x = str(round(info['shapes'][4]['points'][0][0], 2))
          d5y = str(round(info['shapes'][4]['points'][0][1], 2))
          label = x1 + ' ' + y1 + ' ' + w + ' ' + h + ' ' + d1x + ' ' + d1y + ' ' + '0.0' + ' ' + d2x + ' ' + d2y + ' ' + '0.0' + ' ' + d3x + ' ' + d3y + ' ' + '0.0' + ' ' + d4x + ' ' + d4y + ' ' + '0.0' + ' ' + d5x + ' ' + d5y + ' ' + '0.0' + ' ' + '1'

          f.write(label + '\n')

• 标记顺序不一样就改索引
• 最后的1是置信度，因为数据特殊直接写了1

`