facenet代码注释

facenet识别中的embedding代码块

import tensorflow as tf
import numpy as np
import sys
import os
import copy
sys.path.append('../align/')
from MtcnnDetector import MtcnnDetector
from detector import Detector
from fcn_detector import FcnDetector
from model import P_Net,R_Net,O_Net
from utils import *
import config
import cv2
import h5py# In[2]:def main():path='../pictures/embeddings.h5'if os.path.exists(path):print('生成完了别再瞎费劲了！！！')returnimg_arr,class_arr=align_face()# tf.Graph()# 1、它可以通过tensorboard用图形化界面展示出来流程结构# 2、它可以整合一段代码为一个整体存在于一个图中with tf.Graph().as_default():with tf.Session() as sess:#加载模型load_model('../model/')#得到输入与输出tensors#images_placeholder是输入图像的占位符，后面会把images传给它images_placeholder = tf.get_default_graph().get_tensor_by_name("input:0")#embeddings就是卷积网络最后输出的‘‘特征’’embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")#phase_train_placeholder决定了目前是否为训练阶段phase_train_placeholder = tf.get_default_graph().get_tensor_by_name("phase_train:0")keep_probability_placeholder= tf.get_default_graph().get_tensor_by_name('keep_probability:0')# 前向传播计算embeddings# feed_dict的作用是给placeholder创建的tensor赋值的feed_dict = { images_placeholder: img_arr, phase_train_placeholder:False ,keep_probability_placeholder:1.0}embs = sess.run(embeddings, feed_dict=feed_dict)f=h5py.File('../pictures/embeddings.h5','w')class_arr=[i.encode() for i in class_arr]f.create_dataset('class_name',data=class_arr)f.create_dataset('embeddings',data=embs)f.close()# In[3]:
#使用MTCNN网络在原始图片中进行检测和对齐def align_face(path='../pictures/'):# 三个网络的阈值thresh=config.thresh# 设定最小脸大小min_face_size=config.min_face#PNet图片缩小倍数stride=config.stride# 测试选择的网络test_mode=config.test_mode#设detector默认值Nonedetectors=[None,None,None]# 模型放置位置model_path=['../align/model/PNet/','../align/model/RNet/','../align/model/ONet']batch_size=config.batches#对于PNet，其FcnDetector用于识别单张图片PNet=FcnDetector(P_Net,model_path[0])detectors[0]=PNet# 对于RNet，ONet，其Detector用于识别多张图片if test_mode in ["RNet", "ONet"]:RNet = Detector(R_Net, 24, batch_size[1], model_path[1])detectors[1] = RNetif test_mode == "ONet":ONet = Detector(O_Net, 48, batch_size[2], model_path[2])detectors[2] = ONetmtcnn_detector = MtcnnDetector(detectors=detectors, min_face_size=min_face_size,stride=stride, threshold=thresh)#选用图片img_paths=os.listdir(path)# 获取图片类别和路径class_names=[a.split('.')[0] for a in img_paths]img_paths=[os.path.join(path,p) for p in img_paths]scaled_arr=[]class_names_arr=[]for image_path,class_name in zip(img_paths,class_names):#cv2.imread读入图片img = cv2.imread(image_path)#cv2.imshow()函数可以在窗口中显示图像，参数分别为窗口名字和图像#cv2.waitkey()为键盘绑定函数，参数表示等待毫秒数，0表示无期限等待键盘输入
#         cv2.imshow('',img)
#         cv2.waitKey(0)try:boxes_c,_=mtcnn_detector.detect(img)except:print('识别不出图像:{}'.format(image_path))continue#人脸框数量num_box=boxes_c.shape[0]if num_box>0:det=boxes_c[:,:4]det_arr=[]#原图片大小img_size=np.asarray(img.shape)[:2]if num_box>1:#如果保留一张脸，但存在多张，只保留置信度最大的score=boxes_c[:,4]index=np.argmax(score)det_arr.append(det[index,:])else:# 只有一个人脸框的话，那就没得选了det_arr.append(np.squeeze(det))for i,det in enumerate(det_arr):det=np.squeeze(det)#边界框周围的裁剪边缘，获得左上角和右下角的坐标bb=[int(max(det[0],0)), int(max(det[1],0)), int(min(det[2],img_size[1])), int(min(det[3],img_size[0]))]#截取图片cropped = img[bb[1]:bb[3],bb[0]:bb[2],:]#图片截成160*160大小以便作为facenet的输入，并归一化处理scaled =cv2.resize(cropped,(160, 160),interpolation=cv2.INTER_LINEAR)-127.5/128.0scaled_arr.append(scaled)class_names_arr.append(class_name)else:print('图像不能对齐 "%s"' % image_path)scaled_arr=np.asarray(scaled_arr)class_names_arr=np.asarray(class_names_arr)return scaled_arr,class_names_arr# In[4]:def load_model(model_dir,input_map=None):'''重载模型'''ckpt = tf.train.get_checkpoint_state(model_dir)                         saver = tf.train.import_meta_graph(ckpt.model_checkpoint_path +'.meta')   saver.restore(tf.get_default_session(), ckpt.model_checkpoint_path)# In[ ]:if __name__=='__main__':main()

test部分代码块

import tensorflow as tf
import numpy as np
import sys
import os
import copy
from embeddings import load_model
sys.path.append('../align/')
from MtcnnDetector import MtcnnDetector
from detector import Detector
from fcn_detector import FcnDetector
from model import P_Net,R_Net,O_Net
import config
import cv2
import h5py
# 识别人脸阈值
THRED=0.002# In[2]:def main():#读取对比图片的embeddings和class_namef=h5py.File('../pictures/embeddings.h5','r')class_arr=f['class_name'][:]class_arr=[k.decode() for k in class_arr]emb_arr=f['embeddings'][:]# 看是读取摄像头数据还是图片# 读取摄像头,0为摄像头索引，当有多个摄像头时，从0开始编号cap=cv2.VideoCapture(0)# cv2.VideoWriter()指定写入视频帧编码格式，fourcc(*'XVID')设置视频的解码器fourcc = cv2.VideoWriter_fourcc(*'XVID')path='../output'if not os.path.exists(path):os.mkdir(path)out = cv2.VideoWriter(path+'/out.mp4' ,fourcc,10,(640,480))#加载MTCNN模型mtcnn_detector=load_align()with tf.Graph().as_default():with tf.Session() as sess:load_model('../model/')images_placeholder = tf.get_default_graph().get_tensor_by_name("input:0")embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")phase_train_placeholder = tf.get_default_graph().get_tensor_by_name("phase_train:0")keep_probability_placeholder= tf.get_default_graph().get_tensor_by_name('keep_probability:0')while True:#t=GetTickCount()求一段代码的运行时间，单位毫秒ms，返回的时间。t1=cv2.getTickCount()# 捕获一帧图像# 第一个参数ret 为True 或者False,代表有没有读取到图片# 第二个参数frame表示截取到一帧的图片ret,frame = cap.read()if ret == True:img,scaled_arr,bb_arr=align_face(frame,mtcnn_detector)if scaled_arr is not None:feed_dict = { images_placeholder: scaled_arr, phase_train_placeholder:False ,keep_probability_placeholder:1.0}embs = sess.run(embeddings, feed_dict=feed_dict)face_num=embs.shape[0]face_class=['Others']*face_numfor i in range(face_num):diff=np.mean(np.square(embs[i]-emb_arr),axis=1)min_diff=min(diff)print(min_diff)#小于阈值则归为一类，同一个人if min_diff<THRED:index=np.argmin(diff)face_class[i]=class_arr[index]#t=GetTickCount()求一段代码的运行时间，单位毫秒ms，返回的时间。t2=cv2.getTickCount()# getTickFrequency()此函数返回每秒内时钟的周期数t=(t2-t1)/cv2.getTickFrequency()#fps是指画面每秒传输帧数fps=1.0/tfor i in range(face_num):bbox=bb_arr[i]# cv2.putText()这个函数是opencv里面向图像上添加文本内容的函数,各参数依次是：图片，添加的文字，左上角坐标，字体，字体大小，颜色，字体粗细cv2.putText(img, '{}'.format(face_class[i]), (bbox[0], bbox[1] - 2), cv2.FONT_HERSHEY_SIMPLEX,0.5,(0, 255, 0), 2)#画fps值cv2.putText(img, '{:.4f}'.format(t) + " " + '{:.3f}'.format(fps), (10, 20),cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 255), 2)else:img=frame#保存图像a = out.write(img)# cv2.imshow()函数可以在窗口中显示图像，参数分别为窗口名字和图像cv2.imshow("result", img)# waitKey(1)的数字代表等待按键输入之前的无效时间，单位为毫秒，在这个时间段内按键 ‘q’ 不会被记录，在这之后按键才会被记录，并在下一次进入if语段时起作用。也即经过无效时间以后，检测在上一次显示图像的时间段内按键 ‘q’ 有没有被按下，若无则跳出if语句段，捕获并显示下一帧图像。#若此参数置零，则代表在捕获并显示了一帧图像之后，程序将停留在if语句段中一直等待 ‘q’ 被键入。#cv2.waitKey(1)与 0xFF（11111111）相与是因为cv2.waitKey(1)的返回值不止8位，但是只有后8位实际有效，为避免产干扰，通过 ‘与’ 操作将其余位置0。if cv2.waitKey(1) & 0xFF == ord('q'):breakelse:break#关闭读取操作，写入操作和相应的显示窗口。cap.release()out.release()cv2.destroyAllWindows()# In[3]:def load_align():thresh=config.threshmin_face_size=config.min_facestride=config.stridetest_mode=config.test_modedetectors=[None,None,None]# 模型放置位置model_path=['../align/model/PNet/','../align/model/RNet/','../align/model/ONet']batch_size=config.batchesPNet=FcnDetector(P_Net,model_path[0])detectors[0]=PNetif test_mode in ["RNet", "ONet"]:RNet = Detector(R_Net, 24, batch_size[1], model_path[1])detectors[1] = RNetif test_mode == "ONet":ONet = Detector(O_Net, 48, batch_size[2], model_path[2])detectors[2] = ONetmtcnn_detector = MtcnnDetector(detectors=detectors, min_face_size=min_face_size,stride=stride, threshold=thresh)return mtcnn_detector# In[4]:def align_face(img,mtcnn_detector):try:boxes_c,_=mtcnn_detector.detect(img)except:print('找不到脸')return None,None,None#人脸框数量num_box=boxes_c.shape[0]bb_arr=[]scaled_arr=[]if num_box>0:det=boxes_c[:,:4]det_arr=[]img_size=np.asarray(img.shape)[:2]for i in range(num_box):det_arr.append(np.squeeze(det[i]))for i,det in enumerate(det_arr):det=np.squeeze(det)bb=[int(max(det[0],0)), int(max(det[1],0)), int(min(det[2],img_size[1])), int(min(det[3],img_size[0]))]cv2.rectangle(img,(bb[0],bb[1]),(bb[2],bb[3]),(0,255,0),2)bb_arr.append([bb[0],bb[1]])cropped = img[bb[1]:bb[3],bb[0]:bb[2],:]scaled =cv2.resize(cropped,(160,160),interpolation=cv2.INTER_LINEAR)scaled=cv2.cvtColor(scaled,cv2.COLOR_BGR2RGB)-127.5/128.0scaled_arr.append(scaled)scaled_arr=np.array(scaled_arr)return img,scaled_arr,bb_arrelse:print('找不到脸 ')return None,None,None# In[5]:if __name__=='__main__':main()

本文来自互联网用户投稿，文章观点仅代表作者本人，不代表本站立场，不承担相关法律责任。如若转载，请注明出处。 如若内容造成侵权/违法违规/事实不符，请点击【内容举报】进行投诉反馈！