使用caffe python接口进行网络测试，并可视化网络中间内容

import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline


##加载caffe

import sys
import os
import caffe

caffe_root = '/home/hualong/ssd/software/caffe'
sys.path.insert(0, caffe_root + 'python')


##加载网络
USE_GPU = False
if USE_GPU:
    caffe.set_device(0)
    caffe.set_mode_gpu()
else:
    caffe.set_mode_cpu()
    
network_path = '/home/hualong/ssd/coding/my_paper_code/airsim1-zurich/rgb-train/forest_net_deploy.prototxt'    
model_weights = '/home/hualong/ssd/coding/my_paper_code/airsim1-zurich/rgb-train/caffe_model__iter_30000.caffemodel'
labels_file = '/home/hualong/ssd/datasets/airsim_dataset/frame-data/labels_map.txt'

net = caffe.Net(network_path, #定义模型结构 
               model_weights, #包含模型训练权重
               caffe.TEST)    #使用测试模式(训练中不能执行dropout)

## 进行图像预处理

##　图像均值文件的转化　binaryproto＝＝>numpy格式
def convert_mean(binMean,npyMean):
    blob = caffe.proto.caffe_pb2.BlobProto()
    bin_mean = open(binMean, 'rb' ).read()
    blob.ParseFromString(bin_mean)
    arr = np.array( caffe.io.blobproto_to_array(blob) )
    npy_mean = arr[0]
    np.save(npyMean, npy_mean )
binMean='/home/hualong/ssd/datasets/airsim_dataset/frame-data/LMDB_rgb_class/train_mean.binaryproto' 
npyMean='/home/hualong/ssd/datasets/airsim_dataset/frame-data/LMDB_rgb_class/mean.npy' 
convert_mean(binMean,npyMean)
sub_mean = np.load(npyMean).mean(1).mean(1)
print("mean-subtracted values:", zip('BGR',sub_mean))

transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_transpose('data', (2,0,1))  # 图像通道变换，例如从(530,800,3) 变成 (3,530,800)
transformer.set_mean('data', np.load(npyMean).mean(1).mean(1)) #如果你在训练模型的时候没有对输入做mean操作，那么这边也不需要
transformer.set_raw_scale('data', 255)  # 像素值rescale from [0, 1] to [0, 255]
transformer.set_channel_swap('data', (2, 1, 0))  # 通交换通道 from RGB to BGR


## 输入图片，进行前向推理
#设置输入图像大小
net.blobs['data'].reshape(16,       #尽管只检测一张图片，batch size仍为16
                          3,        #3通道
                          180, 320) #图片尺寸227x227

#加载图片，函数声明为load_image(filename, color=True)
image = caffe.io.load_image('/home/hualong/ssd/datasets/airsim_dataset/frame-data/009/rgb_images/mc/img_0_0_1544172750893399000.png') 
#按照之前设置进行预处理
transformed_image = transformer.preprocess('data', image) 
plt.imshow(image) #显示图片

#将图像数据拷贝到为net分配的内存中
net.blobs['data'].data[...] = transformed_image 

#前向传播，进行分类，forward函数说明放到博客最后
#前向传播，跑一遍网络，默认结果为最后一层的blob（也可以指定某一中间层），赋给output
output = net.forward() 

output_prob = output['prob'][0]  #取batch中第一张图像的概率值
output_reference = np.round(output_prob, 3) ##浮点型数保留三位精度
print 'predicted result is:', output_prob
print 'predicted result is:', output_reference
print 'predicted class is:', output_prob.argmax()

labels = np.loadtxt(labels_file, str, delimiter='\t')
print 'output label:{0}  certainty:{1} ' .format(labels[output_prob.argmax()], output_reference[output_prob.argmax()])


top_inds = output_prob.argsort()[::-1][:3]  #逆序排列，取前三个最大值
print 'probabilities and labels:'
zip(output_prob[top_inds], labels[top_inds])