记录一下怎么使用pycaffe调用已有的网络模型识别人脸（物体）

我的哲学原理：比较喜欢从结果向前推，有了能做什么、再去学怎么做？

今天就来看看怎么从图片中识别出人脸：

代码很简单，直接上码：

# -*- coding: utf-8 -*
import numpy as np  
import sys,os  
import cv2
caffe_root = 'E:/bigdata/workspace/caffe-ssd-microsoft/'
sys.path.insert(0, caffe_root + 'python')  
import caffe  
import time;  


net_file= 'faceboxes_deploy.prototxt'  
caffe_model='FaceBoxes_1024x1024.caffemodel'  
test_dir = "images"

#
if not os.path.exists(caffe_model):
    print("FaceBoxes_deploy.caffemodel does not exist,")
    print("use merge_bn.py to generate it.")
    exit()
#
caffe.set_mode_cpu()#gpu
net = caffe.Net(net_file,caffe_model,caffe.TEST) #加载network和model

#分类类别
CLASSES = ('background',
           'face')

#设定图片的shape格式(1,3,1024,1024)
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape}) 
transformer.set_transpose('data', (2,0,1)) #改变维度的顺序，由原始图片(H,W,C)变为(C,H,W)
transformer.set_mean('data', np.array([104,117,123]))  # mean pixel
#transformer.set_raw_scale('data', 255);    # 缩放到[0,255]之间
#transformer.set_channel_swap('data', (2,1,0));   #交换通道，将图片由RGB变为BGR

def preprocess(src):
    img = cv2.resize(src, (1024,1024))
    img = img - 127.5
    img = img * 0.007843
    return img

def postprocess(img, out):   
    h = img.shape[0]
    w = img.shape[1]
    box = out['detection_out'][0,0,:,3:7] * np.array([w, h, w, h])

    cls = out['detection_out'][0,0,:,1]
    conf = out['detection_out'][0,0,:,2]
    return (box.astype(np.int32), conf, cls)

def detect_ok(imgfile):
    
    frame = cv2.imread(imgfile)
    print(frame.shape, frame.ndim, frame.size,frame.dtype, frame.itemsize)
    if frame.ndim<0:
        return False
    #cv2.imshow("org", frame)
    #cv2.waitKey(1000)
    
    # image shape info
    #rows, cols, channels = frame.shape
    height, width, channels = frame.shape
    res = cv2.resize(frame, (1024, 1024), 0.0, 0.0, interpolation=cv2.INTER_CUBIC)
    #cv2.imshow("org", res)
    #cv2.waitKey(1000)
    
    #转成caffe能识别的图片格式
    #res=cv2.cvtColor(res,cv2.COLOR_BGR2RGB)
    res=res/255.
    
    # net input shape info    
    print(net.blobs['data'].data.shape)#input 1,3,1024,1024
    transformed_image = transformer.preprocess('data', res)#frame      
    #cv2.imshow("org", transformed_image)
    #cv2.waitKey(1000)
    
    # 执行上面设置的图片预处理操作，并将图片载入到blob中
    net.blobs['data'].data[...] = transformed_image
    
    time_start=time.time()
    out = net.forward() #运行前向网络预测
    time_end=time.time()
    print(time_end-time_start,"s") 
    
    #显示结果
    box, conf, cls = postprocess(frame, out)
    
    for i in range(len(box)):
        p1 = (box[i][0], box[i][1])
        p2 = (box[i][2], box[i][3])
        cv2.rectangle(frame, p1, p2, (0,255,0), 2)
        p3 = (max(p1[0], 15), max(p1[1], 15))
        title = "%s:%.2f" % (CLASSES[int(cls[i])], conf[i])
        cv2.putText(frame, title, p3, cv2.FONT_ITALIC, 0.6, (0, 255, 0), 2)
    #
    frame = cv2.resize(frame, (int(width/2), int(height/2)), interpolation=cv2.INTER_CUBIC)
    cv2.imshow("org", frame)
    cv2.waitKey(2000)
    return True
     
    

#test here!!!
for f in os.listdir(test_dir):
    if detect_ok(test_dir + "/" + f) == False:
       break

模型文件参见：

https://github.com/zeusees/FaceBoxes