COCO-WholeBody 注释详解

官方数据集说明：https://github.com/jin-s13/COCO-WholeBody

1. 如何预览 json 文件？

使用 Vscode，右键 格式化文档即可。

2. 字段解读

2.1 关键字段之 "images"

"images": [
    {
        "license": 4,
        "file_name": "000000397133.jpg",
        "coco_url": "http://images.cocodataset.org/val2017/000000397133.jpg",
        "height": 427,
        "width": 640,
        "date_captured": "2013-11-14 17:02:52",
        "flickr_url": "http://farm7.staticflickr.com/6116/6255196340_da26cf2c9e_z.jpg",
        "id": 397133 # 图像名字中的数字
    },
	...

上述 "images" 字段的个数是 图像的个数。

2.2 关键字段之 "annotations"

"segmentation": [[实例分割的点]],
"num_keypoints": 10, # 有效的标注的身体关键点， (v>0), (e.g. crowds and small objects, will have num_keypoints=0).
"area": 47803.27955, # 身体的 bbox 的面积
"iscrowd": 0,
"keypoints": [17*3], # (x,y,v) # a visibility flag v defined as v=0: not labeled (in which case x=y=0), v=1: labeled but not visible, and v=2: labeled and visible. A keypoint is considered visible if it falls inside the object segment.
"image_id": 425226,  # 对应 images 中的 id
"bbox": [], # 身体的 bbox，（x,y,w,h） 左上角，宽高
"category_id": 1, # 物体的类别
"id": 183126, # 关键：目标的唯一编号
"face_box": [], 
"lefthand_box": [],
"righthand_box": [],
"lefthand_kpts": [21*3]
"righthand_kpts": [
	308.0,
	239.0,
	0.1408521831035614,
	...
], # 21*3 (xyv) hand 的 v 的含义与 其他的关键点不同点在于：v 有时候是 小于1 的小数，v>0 means that the keypoint is reliable.
"face_kpts": [68*3], #（x,y,v）v 含义同 body 关键点
# The validity of the face/hand/foot are used to minimize the labelling uncertainty. Only if the face/hand images are clear enough for keypoint labeling (for annotators), the validity is True, otherwise False. Invalid cases may include severely blur or occlusion. We only label keypoints/boxes for valid cases. Invalid boxes/keypoints are simply set as all-zero arrays.
"face_valid": false,
"lefthand_valid": true,
"righthand_valid": true,
"foot_valid": true,
"foot_kpts": [6*3] #（x,y,v）v 含义同 body 关键点

简记为

annotation{
 
"face_box": list([x, y, w, h]),
"lefthand_box": list([x, y, w, h]),
"righthand_box": list([x, y, w, h]),
 
"foot_kpts": list([x, y, v] * 6),
"face_kpts": list([x, y, v] * 68),
"lefthand_kpts": list([x, y, v] * 21),
"righthand_kpts": list([x, y, v] * 21),
 
"face_valid": bool,
"lefthand_valid": bool,
"righthand_valid": bool,
"foot_valid": bool,
 
"[cloned]": ...,
}
 
categories[{
"[cloned]": ...,
}]

参考：

1. https://gitee.com/mirrors_jin-s13/COCO-WholeBody/blob/master/data_format.md