前言
人脸关键点检测是计算机视觉中的重要任务，它在人脸识别、表情识别、人脸美化等应用中发挥着关键作用。然而，为了训练高效准确的人脸关键点检测模型，我们需要大量的标注数据集。在本篇博客中，我们将介绍如何快速制作高质量的人脸关键点数据集，以支持后续模型的训练与优化。
关键点标注信息展示
在这里向大家介绍一款简单常见的关键点标注程序：labelme，大家可以通过这款软件进行关键点标注。在这里我以帅气的郭富城为例子进行标注五官关键点以及相应的json文件：
shell
复制代码
{ "version": "5.2.1", "flags": {}, "shapes": [ { "label": "0", "points": [ [ 74.02476780185759, 130.80495356037153 ] ], "group_id": null, "description": "", "shape_type": "point", "flags": {} }, { "label": "1", "points": [ [ 126.656346749226, 136.687306501548 ] ], "group_id": null, "description": "", "shape_type": "point", "flags": {} }, { "label": "2", "points": [ [ 101.26934984520125, 169.1950464396285 ] ], "group_id": null, "description": "", "shape_type": "point", "flags": {} }, { "label": "3", "points": [ [ 101.57894736842104, 194.89164086687308 ] ], "group_id": null, "description": "", "shape_type": "point", "flags": {} }, { "label": "4", "points": [ [ 45.8513931888545, 134.82972136222912 ] ], "group_id": null, "description": "", "shape_type": "point", "flags": {} }, { "label": "5", "points": [ [ 162.26006191950466, 143.49845201238392 ] ], "group_id": null, "description": "", "shape_type": "point", "flags": {} } ], "imagePath": "300.jpg", "imageData": "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", "imageHeight": 300, "imageWidth": 240
标注信息获取
分析【关键点标注信息展示】的信息，我们可以得到其组成为version 、 flags 、 shapes 、 imagePath 、 imageData 、 imageHeight 、 imageWidth 。 其中的【version 、 flags】 为固定信息，【shapes】中的信息为标注的关键点信息 ， 【imagePath 、 imageData 、 imageHeight 、imageWidth】分别为图像名称 、 图像Base64编码字符串 、 图像的高 、 图像的宽。而在【shapes】中的信息仅label和points是变动的，分布代表类别和坐标点。
关键点检测
既然需要制作人脸关键点数据集，所以人脸关键点的检测是必不可少的，我们仍然是通过开源项目进行人脸关键点检测。在这里我们选择的是人脸关键点较少的开源项目检测：【mediapipe】在这里大家可以翻阅我往期关于【mediapipe】介绍。例如：
- 【特效】对实时动态人脸进行马赛克及贴图马赛克处理及一些拓展
- 【姿态估计】从理论到实践逐步分析讲解传统姿态估计算法
- 【实操：人脸矫正】两次定位操作解决人脸矫正问题
- 一起来学MediaPipe（二）人脸面网格
- 一起来学MediaPipe（二）人脸面网格
在这一步骤中我们核心目的是为了使用mediapipe包对人脸关键点检测后输出：
shell
复制代码
import cv2 import mediapipe as mp mp_face_detection = mp.solutions.face_detection mp_drawing = mp.solutions.drawing_utils def get_face_info(image): image.flags.writeable = False image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) results = face_detection.process(image) facial_axis = None if results.detections: for detection in results.detections: facial_axis = detection.location_data.relative_keypoints return facial_axis def AxisTransformation(w, h, img_fa): # 左眼 left_eye = [img_fa[0].x * h, img_fa[0].y * w] # 右眼 right_eye = [img_fa[1].x * h, img_fa[1].y * w] # 鼻子 nose = [img_fa[2].x * h, img_fa[2].y * w] # 嘴巴 mouth = [img_fa[3].x * h, img_fa[3].y * w] # 左耳 left_ear = [img_fa[4].x * h, img_fa[4].y * w] # 右耳 right_ear = [img_fa[5].x * h, img_fa[5].y * w] points = [[left_eye], [right_eye], [nose], [mouth], [left_ear], [right_ear]] labels = ["left_eye", "right_eye", "nose", "mouth", "left_ear", "right_ear"] return points, labels
image信息获取
通过关键点标注分析我们可以获得在image信息中需要填入的信息，其中图像的尺寸和名称较为轻易获取，我们在批量读取图像的时候通过for循环可以解决，在这里我们着重解决BASE64编码字符串的获取问题：
shell
复制代码
def get_imageData(img): # 将图像数据转换为Base64编码字符串 image_data_binary = cv2.imencode(".jpg", img)[1].tobytes() image_data_base64 = base64.b64encode(image_data_binary).decode() return image_data_base64
shapes信息获取
由于一张图内的关键点往往不是一个，是需要多个关键点组合而成的，我们从上述的分析中也可以得到shapes信息中改变的也仅仅为lebel和points信息，而group_id、description、shape_type和flags参数爆出不变。同时每个关键点的信息也都是有如下关键信息所示，我们可以这样定义一个函数进行获取shapes信息：
shell
复制代码
def get_shapes_data(points, label): info = { "label": label, "points": points, "group_id": None, "description": "", "shape_type": "point", "flags": {} } return info
生成标注文件
在批量生成标注数据时，我们假定是通过文件夹中的图像进行处理，基础逻辑流程为：
可以通过下面这段代码实现上述逻辑，我们使用四大天王的照片进行测试实验：
shell
复制代码
if __name__ == "__main__": img_path = "./img/" with mp_face_detection.FaceDetection(model_selection=0, min_detection_confidence=0.5) as face_detection: for name in os.listdir(img_path): src_img = cv2.imread(img_path + name) # 读取原始图像 facial_axis = get_face_info(src_img) # 五官坐标集 if facial_axis is not None: # 判定关键点坐标是否存在 Height, Width, _ = src_img.shape imageData = get_imageData(src_img) # 获取图像的imageData信息 # 标签的基础信息 json_base = { "version": "5.2.1", "flags": {}, "shapes": None, "imagePath": name, "imageData": imageData, "imageHeight": Height, "imageWidth": Width } points, labels = AxisTransformation(Height, Width, facial_axis) # 获取关键点坐标， 类别名称 # 获取关键点信息 shape_infos = [] for epoch in range(len(points)): shape_info = get_shapes_data(points[epoch], labels[epoch]) shape_infos.append(shape_info) # 对基础款进行修改 json_base["shapes"] = shape_infos json.dump(json_base, open('./img/%s.json' % name[:-4], 'w'), indent=2)
生成后观察文件夹下图像及标注文件：
使用labelme检验标注结果：
结语
周五了，匆忙中实现了一个人脸关键点数据集的快速制作脚本，仅以此篇以飨读者！这样大家可以快速制作自己的关键点数据集了，大家也可以使用Dlib包制作自己的人脸关键点或者其他开源项目制作自己的关键点数据集。
