deepface：人脸检测、人脸识别、人脸验证笔记

https://github.com/serengil/deepface

pip install deepfacepip install tf-keraspip install facenet-pytorch

$ git clone https://github.com/serengil/deepface.git$ cd deepface$ pip install -e .

from deepface import DeepFace

backends = [  'opencv',   'ssd',   'dlib',   'mtcnn',   'fastmtcnn',  'retinaface',   'mediapipe',  'yolov8',  'yunet',  'centerface',]alignment_modes = [True, False]

obj = DeepFace.verify(  img1_path = "tests\dataset\img1.jpg",   img2_path = "tests\dataset\img2.jpg",   detector_backend = backends[0],  align = alignment_modes[0],)

img1_path: 第一张图片的路径。img2_path: 第二张图片的路径。model_name: 使用的人脸识别模型的名称，默认为"VGG-Face"。detector_backend: 使用的检测器后端的名称，默认为"opencv"。distance_metric: 用于计算相似度的度量方法，默认为"cosine"。enforce_detection: 是否强制检测人脸，默认为True。align: 是否对齐人脸，默认为True。normalization: 归一化方法，默认为"base"。

{  'verified': True,  'distance': 0.412801693879869,  'threshold': 0.68,  'model': 'VGG-Face',  'detector_backend': 'opencv',  'similarity_metric': 'cosine',  'facial_areas': {    'img1': {      'x': 339,      'y': 218,      'w': 768,      'h': 768,      'left_eye': (850, 524),      'right_eye': (571, 517)    },    'img2': {      'x': 524,      'y': 201,      'w': 491,      'h': 491,      'left_eye': (858, 388),      'right_eye': (663, 390)    }  },  'time': 2.27}

verified: 布尔值，表示两张图片是否属于同一个人。如果距离小于等于阈值，则返回True，否则返回False。distance: 浮点数，表示两张图片之间的距离。threshold: 浮点数，表示用于判断两张图片是否属于同一个人的阈值。model: 字符串，表示使用的人脸识别模型的名称。detector_backend: 字符串，表示使用的检测器后端的名称。similarity_metric: 字符串，表示用于计算相似度的度量方法。facial_areas: 字典，包含两个键值对，分别表示两张图片的人脸区域信息。键为"img1"和"img2"，值为对应的人脸区域数据。time: 浮点数，表示执行该段代码所花费的时间，保留两位小数。

models = [  "VGG-Face",   "Facenet",   "Facenet512",   "OpenFace",   "DeepFace",   "DeepID",   "ArcFace",   "Dlib",   "SFace",  "GhostFaceNet",]

dfs = DeepFace.find(  img_path = "img1.jpg",  db_path = "E:\lfw-funneled\lfw_funneled",   model_name = models[1],)

img_path：要查找的图像的路径。db_path：包含人脸数据库的文件夹路径。model_name：使用的人脸识别模型，默认为"VGG-Face"。distance_metric：用于计算相似度的距离度量方法，默认为"cosine"。enforce_detection：是否强制检测人脸，默认为True。detector_backend：用于检测人脸的后端，默认为"opencv"。align：是否对齐人脸，默认为True。normalization：归一化方法，默认为"base"。silent：是否在执行过程中保持静默，默认为False。

identity:表示最相似图像的身份或标识。verified:表示是否找到了匹配的图像。如果为 True，则表示找到了匹配；如果为 False，则表示未找到匹配。distance:表示查询图像与找到的最相似图像之间的距离。

http://vis-www.cs.umass.edu/lfw/

https://www.kaggle.com/datasets/atulanandjha/lfwpeople

img_path：输入图像的路径。target_size：目标大小，默认为(224, 224)。detector_backend：检测器后端，默认为"opencv"。enforce_detection：是否强制检测人脸，默认为True。align：是否对齐人脸，默认为True。grayscale：是否将图像转换为灰度图，默认为False。

face_objs = DeepFace.extract_faces(  img_path = "img.jpg",   detector_backend = backends[4],  align = alignment_modes[0],)

'facial_area': {'x': 1145, 'y': 802, 'w': 99, 'h': 127, 'left_eye': (1219, 856), 'right_eye': (1174, 855)}, 'confidence': 1.0}

import cv2 as cvimg = cv.imread('C:\Users\DELL\Desktop\10.jpg',1)cv.imshow('image',img)

face_objs = DeepFace.extract_faces(img_path = img,detector_backend = backends[4],align = alignment_modes[0],)for item in face_objs:    rect_img1 = item['facial_area']    cv.rectangle(img, (rect_img1["x"], rect_img1["y"]),(rect_img1["x"] + rect_img1["w"], rect_img1["y"] + rect_img1["h"]),(0, 255, 0),1)cv.imshow('image',img)

img = cv.imread('C:\Users\DELL\Desktop\17.jpg',1)face_objs = DeepFace.extract_faces(img_path = img,detector_backend = backends[4],align = alignment_modes[0],)for item in face_objs:    rect_img1 = item['facial_area']    cv.rectangle(img, (rect_img1["x"], rect_img1["y"]),(rect_img1["x"] + rect_img1["w"], rect_img1["y"] + rect_img1["h"]),(0, 255, 0),1)

import numpy as npimport cv2 as cvfrom deepface import DeepFacebackends = [  'opencv',   'ssd',   'dlib',   'mtcnn',   'fastmtcnn',  'retinaface',   'mediapipe',  'yolov8',  'yunet',  'centerface',]alignment_modes = [True, False]cap = cv.VideoCapture("C:\Users\DELL\Desktop\109.mp4")fourcc = cv.VideoWriter_fourcc(*'XVID')# 获取视频帧的宽度和高度frame_width = int(cap.get(cv.CAP_PROP_FRAME_WIDTH))frame_height = int(cap.get(cv.CAP_PROP_FRAME_HEIGHT))out = cv.VideoWriter('C:\Users\DELL\Desktop\output.mp4', fourcc, 20.0, (frame_width, frame_height))while cap.isOpened():    ret, frame = cap.read()    if not ret:        print("Can't receive frame (stream end?). Exiting ...")        break    face_objs = DeepFace.extract_faces(img_path=frame, detector_backend=backends[4], align=alignment_modes[0], enforce_detection=False)    if face_objs:  # 检查是否检测到人脸        for item in face_objs:            rect_img1 = item['facial_area']            cv.rectangle(frame, (rect_img1["x"], rect_img1["y"]),                         (rect_img1["x"] + rect_img1["w"], rect_img1["y"] + rect_img1["h"]),                         (0, 255, 0), 2)    out.write(frame)# 释放资源cap.release()out.release()cv.destroyAllWindows()