python-opencv：基于SSD的人脸检测

SSD：

SSD是一种基于深度学习的目标检测算法，opencv在3.3版本以后将其引入作为基于深度学习的人脸检测器；

• 模型

opencv实现的SSD人脸检测器的骨干网络是REsNet-10，当前它提供了两个训练好的模型：基于深度学习框架caffe训练的模型和基于TensorFlow训练的模型

下载地址：https://github.com/opencv/opencv/blob/master/samples/dnn/face_detector/weights.meta4

• 配置文件

下载地址：

https://github.com/opencv/opencv/tree/master/samples/dnn/face_detector

face_detector文件分析：
deploy.prototxt：调用.caffemodel时的测试网络文件
how_to_train_face_detector.txt：如何使用自定义数据集来训练网络的说明
solver.prototxt：超参数文件
test.prototxt：测试网络文件
train.prototxt：训练网络文件

• 模型导入方式

（1）基于深度学习框架caffe训练的模型导入

1. config_file = "face_detector/deploy.prototxt";

2. model_file = "face_detector/res10_300x300_ssd_iter_140000_fp16.caffemodel";

3. net = cv2.dnn.readNetFromCaffe(config_file, model_file)

（2）基于TensorFlow训练的模型导入

1. config_file = "face_detector/opencv_face_detect.pbtxt";

2. model_file = "face_detector/opencv_face_detect_uint8.pb";

3. net = cv2.dnn.readNetFromTensorflow(model_file, config_file)

完整测试示例： 

# -*- coding: utf-8 -*-
"""
Created on Sun Jun 16 09:27:01 2019

@author: zfjua
"""

import cv2 as cv
from cv2 import dnn

net = dnn.readNetFromCaffe('./data/face_detector/deploy.prototxt', './data/face_detector/res10_300x300_ssd_iter_140000_fp16.caffemodel')


inWidth = 300
inHeight = 300
confThreshold = 0.5

frame = cv.imread('.\\image\\test.png')
cols = frame.shape[1]
rows = frame.shape[0]

net.setInput(dnn.blobFromImage(frame, 1.0, (inWidth, inHeight), (104.0, 177.0, 123.0), False, False))
detections = net.forward()
# perf_stats = net.getPerfProfile()
# print('Inference time, ms: %.2f' % (perf_stats[0] / cv.getTickFrequency() * 1000))
for i in range(detections.shape[2]):
    confidence = detections[0, 0, i, 2]
    if confidence > confThreshold:
        xLeftBottom = int(detections[0, 0, i, 3] * cols)
        yLeftBottom = int(detections[0, 0, i, 4] * rows)
        xRightTop = int(detections[0, 0, i, 5] * cols)
        yRightTop = int(detections[0, 0, i, 6] * rows)

        cv.rectangle(frame, (xLeftBottom, yLeftBottom), (xRightTop, yRightTop), (0, 0, 255))

cv.imshow("detections", frame)
cv.waitKey(0)

cv.destroyAllWindows()