自己数据集转化为coco数据集

1.将图片copy到对应文件夹

%cd /home/aistudio
!cp -r data/hand143_panopticdb/imgs/* data/coco/images/

2.提取标签数据

import json
import os
import cv2
dataset = {'info':[],'categories':[],'images':[],'annotations':[]}
# 根路径，里面包含images(图片文件夹)，annos.txt(bbox标注)，classes.txt(类别标签),以及annotations文件夹(如果没有则会自动创建，用于保存最后的json)
data_path = 'data/hand143_panopticdb/hands_v143_14817.json'
root_path = 'data/coco'#保存路径
# 用于创建训练集或验证集v
phase = 'train'

f = open(data_path,encoding='utf-8')
content = json.load(f)
# print(content['root'][0])
for j in range(len(content['root'])):
    keypoints=[]
    content['root'][j]['img_paths'] = os.path.join(root_path,content['root'][j]['img_paths'])#将路径补全
        
    
    dataset['images'].append({'id': j,
     'width':content['root'][j]['img_width'], 
     'height':content['root'][j]['img_height'] ,
     'file_name':content['root'][j]['img_paths'],
     'license':None,
     'flickr_url':None,
     'coco_url':None,
     'date_captured':'2020/5/1'})
    for i in range(len(content['root'][j]['joint_self'])):
        keypoints=keypoints+content['root'][j]['joint_self'][i]    
    dataset['annotations'].append({"keypoints": keypoints, "num_keypoints": len(content['root'][j]['joint_self']),"[cloned]":None})
    # 建立类别标签和数字id的对应关系
joint=[]
for i in range(len(content['root'][1]['joint_self'])):
    joint.append('joint'+str(i))
dataset['categories'].append({'id': 1, 'name':joint, 'supercategory': 'mark'})


dataset['info'].append({
"description":"This is stable 1.0 version of the 2014 MS COCO dataset.",
"url":"http:\/\/mscoco.org",
"version":"1.0","year":2014,
"contributor":"Microsoft COCO group",
"date_created":"2015-01-27 09:11:52.357475"
})

3.保存结果

# 保存结果的文件夹
folder = os.path.join(root_path, 'annotations')
if not os.path.exists(folder):
  os.makedirs(folder)
json_name = os.path.join(root_path, 'annotations/{}.json'.format(phase))
with open(json_name, 'w') as f:
  json.dump(dataset, f)