YOLO-V5-s 实现人脸框检测

• 下载yolov-v5代码，并安装依赖库（推荐adaconda创建虚拟环境后安装）

  github地址：https://github.com/ultralytics/yolov5

  yolo-v5官方文档：https://docs.ultralytics.com/yolov5/tutorials/train_custom_data/#train-on-custom-data
  

  ?

  1 2 3

  git clone https://github.com/ultralytics/yolov5  # clone cd yolov5 pip install -r requirements.txt  # install

 

 

• 下载人脸检测训练集
  WIDER FACE训练集：http://shuoyang1213.me/WIDERFACE/index.html
  

  

   

  
  
  
  
• 将WINDER FACE人脸框信息转为YOLO格式
  由于训练速度慢，验证集数据量最少，本次将使用验证集作为训练集使用
  将训练集图片和对应人脸框信息放到YOLO-v5项目下
  

  
  
  
  

•  

  使用脚本将WINDER FACE数据转化为YOLO格式

  ?

  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 76 77 78 79 80 81 82 83 84 85 86 87 88

  import os import shutil     class WiderFace(object):     def __init__(self, label_path, img_dir_path, save_path, class_num):         self.class_num = class_num         self.label_path = label_path         self.img_dir_path = img_dir_path         self.save_path = save_path         # self.train_percent = train_percent         # self.val_percent = val_percent         # self.test_percent = test_percent         self.yolo_img_path = os.path.join(save_path, "images")         self.yolo_label_path = os.path.join(save_path, "labels")         if not os.path.exists(self.yolo_img_path ):             os.makedirs(self.yolo_img_path)         if not os.path.exists(self.yolo_label_path ):             os.makedirs(self.yolo_label_path)       def analysis_line(self, line):         """判断当前行数据类型，并返回"""         if "." in line:             img_path = os.path.join(self.img_dir_path, line)             img_path = img_path.replace("\n", "")             return "img", img_path         if " " in line:             # 左、上、宽、高             landmark = line.split(" ")[:4]             return "bbox", landmark         return "box_sum", line.replace("\n", "")       def save_img(self, img_path):         """复制图片到新目录并按照图片数量重新命名"""         file_count = len(os.listdir(self.yolo_img_path))         new_file_name = "%05d" % file_count + os.path.splitext(img_path)[-1]         new_file_path = os.path.join(self.yolo_img_path, new_file_name)         shutil.copy(img_path, new_file_path)         return new_file_name       def save_yolo(self):         with open(self.label_path, "r") as f:             bbox_list = []             cur_img_name = ""             for line in f:                 line_type, data = self.analysis_line(line)                 if line_type == 'img':                     # 显示图片和人脸框                     # if bbox_list:                     #     from PIL import Image, ImageDraw                     #     new_file_path = os.path.join(self.yolo_img_path, cur_img_name)                     #     img = Image.open(new_file_path)                     #     draw = ImageDraw.Draw(img)                     #     for item in bbox_list:                     #         draw.rectangle(xy=item, outline="red")                     #     img.show()                     #     img.close()                     #     bbox_list.clear()                       # 拷贝图片并重新命名                     img_path = os.path.join(self.yolo_img_path, data)                     cur_img_name = self.save_img(img_path)                 elif line_type == 'bbox':                     # 存储所属分类和人脸框信息                     from PIL import Image                     new_file_path = os.path.join(self.yolo_img_path, cur_img_name)                     img = Image.open(new_file_path)                     w, h = img.size                     x_center, y_center = int(data[0])+int(data[2])/2, int(data[1])+int(data[3])/2                     yolo_label = "1 %s %s %s %s\n" % (x_center/w, y_center/h, int(data[2])/w, int(data[3])/h)                     label_path = os.path.join(self.yolo_label_path, cur_img_name.split(".")[0]) + ".txt"                     with open(label_path, "a+") as f:                         f.writelines(yolo_label)                       # 存储人脸框信息                     xy = (int(data[0]), int(data[1]), int(data[0])+int(data[2]), int(data[1])+int(data[3]))                     bbox_list.append(xy)     if __name__ == "__main__":     label_path = "/Users/wb-fcj414969/project/yolov5-master/samples/wider_face/wider_face_val_bbx_gt.txt"     img_dir_path = "/Users/wb-fcj414969/project/yolov5-master/samples/wider_face/WIDER_val/images"     save_dir_path = "/Users/wb-fcj414969/project/yolov5-master/samples/wider_face/yolo/"     # train_percent = 0.6     # val_percent = 0.2     # test_percent = 0.2     wider_face = WiderFace(label_path, img_dir_path, save_dir_path, 1)     wider_face.save_yolo()

  　　

• 新建数据集配置文件face.yaml放入data文件中

  ?

  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

  # YOLOv5 🚀 by Ultralytics, AGPL-3.0 license # COCO128 dataset https://www.kaggle.com/ultralytics/coco128 (first 128 images from COCO train2017) by Ultralytics # Example usage: python train.py --data coco128.yaml # parent # ├── yolov5 # └── datasets #     └── coco128  ← downloads here (7 MB)     # Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..] path:*****/samples/wider_face/yolo  # dataset root dir train: images  # train images (relative to 'path') 128 images val: images  # val images (relative to 'path') 128 images test:  # test images (optional)   # Classes names:   0: person   1: face     # Download script/URL (optional) #download: https://ultralytics.com/assets/coco128.zip

  　　

   

• shell命令启动训练脚本
  python train.py --img 640 --epochs 3 --data face.yaml --weights yolov5s.pt