常用数据集简介

# COCO格式数据集

# 01、COCO 格式数据集标注解析

分为四部分信息

"info": {
    "description": "CowboySuit",
    "url": "http://github.com/dmlc/gluon-cv",
    "version": "1.0",
    "year": 2021,
    "contributor": "GluonCV/AutoGluon",
    "date_created": "2021/07/01"
  	},

1
2
3
4
5
6
7
8

info 主要描述数据的元信息，一般就是标注数据的版权和来源之类的

"images": [
      {
      	"id": 11684018977583795263,
      	"file_name": "a225f95b03ce0c3f.jpg",
      	"neg_category_ids": [
        	120
      	],
      	"pos_category_ids": [
	        69,
    	    216,
        	308,
        	433
      	],
      	"width": 1024,
      	"height": 683,
      	"source": "OpenImages"
   	  },
     ...
  	]

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

images 是一个列表，列表里的每个实例都描述一张图像，图像比较重要的属性值如下
- id
- file_name
- height
- weight

"annotations":[
	  {
      	"id": 9502719,
      	"image_id": 11684018977583795263,
      	"freebase_id": "/m/032b3c",
      	"category_id": 588,
      	"iscrowd": false,
      	"bbox": [
        524.16,
        255.0,
        365.44,
        411.98
      	],
      	"area": 150553.52
   	  }
      ..
  	]

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

annotations 则是对应图像的标注，标注比较重要的属性值如下
- image_id
- category_id
- bbox: [x, y, width, height]
- bbox 的格式是目标框左上角的坐标以及该框的宽和高

"categories": [
    {
        "id": 87,
        "name": "belt",
        "freebase_id": "/m/0176mf"
    },
    ...
]

1
2
3
4
5
6
7
8

categories 则是所有的类别，比较重要的属性值如下
- id
- name

# 02、COCO 格式数据集标注可视化

# 2.1 目标框的标注可视化

# import cv2
import json
import os

import mmcv
import numpy as np
from pycocotools.coco import COCO


def visualization_bbox(num_image, json_path, img_path):  # 需要画的第num副图片， 对应的json路径和图片路径
    with open(json_path) as annos:
        annotation_json = json.load(annos)

    print('the annotation_json num_key is:', len(annotation_json))  # 统计json文件的关键字长度
    print('the annotation_json key is:', annotation_json.keys())  # 读出json文件的关键字
    print('the annotation_json num_images is:', len(annotation_json['images']))  # json文件中包含的图片数量

    label_id_cls_dict = dict()
    for i in range(len(annotation_json['categories'][::])):
        id = annotation_json['categories'][i]['id']
        name = annotation_json['categories'][i]['name']
        label_id_cls_dict[id] = name
    print('the categories id to name dict is', label_id_cls_dict)

    id = annotation_json['images'][num_image - 1]['id']  # 读取图片id
    image_name = annotation_json['images'][num_image - 1]['file_name']  # 读取图片名
    image_path = os.path.join(img_path, str(image_name).zfill(5))  # 拼接图像路径
    image = mmcv.imread(image_path)
    bboxes = []
    labels = []

    for i in range(len(annotation_json['annotations'][::])):
        if annotation_json['annotations'][i]['image_id'] == id:
            x, y, w, h = annotation_json['annotations'][i]['bbox']  # 读取边框
            label_id = annotation_json['annotations'][i]['category_id']
            label = label_id_cls_dict[label_id]

            # 将左上和右下的坐标加入列表
            bboxes.append([int(x), int(y), int(x + w), int(y + h)])
            labels.append(label)
    print('The num_bbox of the display image is:', len(bboxes))
    mmcv.imshow_det_bboxes(image, np.array(bboxes), np.array(labels))


def visualization_bbox_coco_api(num_image, json_path, img_path):
    coco = COCO(json_path)
    img = list(coco.imgs.items())[num_image][1]

    image_id = img['id']  # 读取图像id
    image_name = img['file_name']  # 读取图像名字

    image = mmcv.imread(os.path.join(img_path, image_name))  # 读取图像
    annIds = coco.getAnnIds(imgIds=image_id)
    anns = coco.loadAnns(annIds)

    bboxes = []
    labels = []
    for i in range(len(annIds)):
        x, y, w, h = anns[i]['bbox']  # 读取边框
        label_id = anns[i]['category_id']
        label = coco.cats[label_id]['name']
        bboxes.append([int(x), int(y), int(x + w), int(y + h)])

        labels.append(label)

    print('The num_bbox of the display image is:', len(bboxes))
    mmcv.imshow_det_bboxes(image, np.array(bboxes), np.array(labels))


if __name__ == "__main__":
    train_json = '/home/muyun99/competition/detection/data/kaggle_cowboyoutfits/annotations.json'
    train_path = '/home/muyun99/competition/detection/data/kaggle_cowboyoutfits/images'
    for i in range(10):
        visualization_bbox(i, train_json, train_path)
        visualization_bbox_coco_api(i, train_json, train_path)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75

上次更新: 2021/08/08, 21:56:44

← mmdetection COCO格式数据集 cowfits竞赛记录→