caffe 读取图片，测试

#coding:utf-8
import sys
sys.path.insert(0,"/home/caffe/python")
import caffe
import cv2
import numpy as np
import time
import glob
from tqdm import tqdm

def person_top1():
    deploy = './v0.prototxt'
    caffe_model = './v0.caffemodel'
    caffe.set_device(0)
    caffe.set_mode_gpu()
    net = caffe.Net(deploy,caffe_model,caffe.TEST)   #

    print("net load suceed")
    print("==============================================>")
    print("now extracting img feature, please wait a minutes ...")

    img_feats = []
    label = []
    for path in tqdm(glob.glob("./test_img/*.png")):
        # tensor preprocess
        # print path
        img = cv2.imread(path)
        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)  #bgr to rgb
        img = cv2.resize(img,(128,256))
        img = img.transpose(2, 0, 1)  # hwc   to chw
		# img = (img-127.5)*0.0078125  #normalize 
		
        mean = [0.485, 0.456, 0.406]
        std = [0.229, 0.224, 0.225]
        img = img / 255.0
        # Normalization
        for channel, _ in enumerate(img):
            img[channel] -= mean[channel]
            img[channel] /= std[channel]

        net.blobs['data'].data[...] = img  #nchw 数据输入，但是这里net.blobs['data'].data[...] ，输入chw
        net.forward()  #前向传输也有其他写法
        
        feature = np.float64(net.blobs["batch_norm_blob54"].data[0])
        img_feats.append(feature.flatten())
        lab = path.split("/")[2].split("_")[0]
        label.append(int(lab))

    img_feats = np.array(img_feats).astype(np.float32)  # \u6240\u6709\u56fe\u7247numpy,

    g_feats = img_feats
    t_feats = img_feats
    g_feats = g_feats / np.sqrt(np.sum(g_feats ** 2, -1, keepdims=True))
    t_feats = t_feats / np.sqrt(np.sum(t_feats ** 2, -1, keepdims=True))
    # gallery_label=np.concatenate((test_feats30, test_feats40), axis=0) #\u62fc\u63a5

    ### \u7279\u5f81\u63d0\u53d6\u7ed3\u675f\uff0c\u8fdb\u884c \u6bd4\u5bf9\uff0c top
    print("all img feature dimension", img_feats.shape)
    correct10 = 0
    correct1 = 0
    for i, line in enumerate(t_feats):
        line = np.tile(line, (len(g_feats), 1))  # 一个特征维度，重复 所有维度
        dis = np.sum(g_feats * line, 1)  # save index  correspond index
        sort_index = np.argsort(-dis, axis=0)  #大小排序 没有- 小大排序

        top100 = []
        temp_label = []
        # for j in range(101): #get sort index
        for j in range(11):  # get sort index
            # top100.append(dis[sort_index[j]])
            temp_label.append(label[sort_index[j]])  ##gallery label

        if label[i] in temp_label[1:11]:  # test_label   is 1,because Remove its own
            correct10 = correct10 + 1
        if label[i] in temp_label[1:2]:
            correct1 = correct1 + 1

    print("=====================================================>")
    print ("top1 : ", correct1 / float(len(t_feats)))
    print ("top10 : ", correct10 / float(len(t_feats)))


## test
top1()