1、自定义少量数据
（1）加入图片
对于自定义少量数据，可以通过自己上传的方式，即进入后台启动docker-compose.yml所在的文件夹，能够发现一个外挂的datasets数据集。
进入数据集，里面是建立的各种datasets，创建数据集也很简单，如下：
加入一张图片，图片通过将本地直接拖到服务器上面的指定目录下面即可，即test，刷新就能看到数据了。
（2）创建类别
创建类别，主要是你对图像进行标注的时候，不同区域到底有那些分类，方式如下：
（3）开始标注
点击进入dataset，然后开始标注，第一步，点击右侧的类别，然后第二步用高亮的矩形框标注
（4）下载标注好的数据
标注完成后需要拿到标注好的数据进行使用，点击下载
下载后的标注结果：
{
"images":
[
{
"id": 4,
"dataset_id": 5,
"path": "/datasets/test/3.jpg",
"width": 499,
"height": 333,
"file_name": "3.jpg"
}
],
"categories":
[
{
"id": 82,
"name": "脸",
"supercategory": "",
"color": "#ff0000",
"metadata":
{},
"keypoint_colors":
[]
},
{
"id": 84,
"name": "屁股",
"supercategory": "",
"color": "#56bd0c",
"metadata":
{},
"keypoint_colors":
[]
}
],
"annotations":
[
{
"id": 16,
"image_id": 4,
"category_id": 82,
"segmentation":
[
[
97.3,
187,
97.3,
203,
81.9,
203,
81.9,
187
]
],
"area": 240,
"bbox":
[
82,
187,
15,
16
],
"iscrowd": false,
"isbbox": true,
"color": "#31b4d9",
"keypoints":
[],
"metadata":
{
"slice1": 1
}
},
{
"id": 17,
"image_id": 4,
"category_id": 82,
"segmentation":
[
[
432.4,
136.3,
432.4,
149.4,
417.7,
149.4,
417.7,
136.3
]
],
"area": 182,
"bbox":
[
418,
136,
14,
13
],
"iscrowd": false,
"isbbox": true,
"color": "#0341bb",
"keypoints":
[],
"metadata":
{
"slice1": 1
}
},
{
"id": 18,
"image_id": 4,
"category_id": 84,
"segmentation":
[
[
256.7,
181.7,
256.7,
198.3,
234.8,
198.3,
234.8,
181.7
]
],
"area": 352,
"bbox":
[
235,
182,
22,
16
],
"iscrowd": false,
"isbbox": true,
"color": "#4cf622",
"keypoints":
[],
"metadata":
{
"slice1": 1
}
},
{
"id": 19,
"image_id": 4,
"category_id": 84,
"segmentation":
[
[
92.6,
214.6,
92.6,
229,
75.7,
229,
75.7,
214.6
]
],
"area": 238,
"bbox":
[
76,
215,
17,
14
],
"iscrowd": false,
"isbbox": true,
"color": "#e166c2",
"keypoints":
[],
"metadata":
{
"slice1": 1
}
}
]
}
2、数据集直接按照格式整理
（1）先把图片放到指定位置
（2）上传json文件
{
"info":
{
"description": "img",
"url": "http://cocodataset.org",
"version": "1.0",
"year": 2021,
"contributor": "this is my sample",
"date_created": "2021/08/27"
},
"images":
[
{
"id": 1,
"path": "/datasets/demo/1.jpg",
"width": 1080,
"height": 808,
"file_name": "1.jpg"
}
],
"categories":
[
{
"supercategory": "person",
"id": 91,
"name": "face"
},
{
"supercategory": "person",
"id": 90,
"name": "leg"
}
]
}
如上，只要把该文件里面images里面的path路径写对，就可以了，然后刷新
