获取模型执行过程中的中间结果

   def testImage(self,image_location,image_size):
      mean = self.readMeanFile()
      image = ImageProcess(image_location,image_size,mean).getImage()
      fileOp = FileOperation()
      #1.read the pb file as a string
      init_path = self.MODEL_PATH+"/"+self.MODEL_NAME+"/"+self.INIT_NAME
      #print(init_path)
      init_pb = fileOp.readFile(filePath = init_path ,openMode = "rb")

      predict_path = self.MODEL_PATH+"/"+self.MODEL_NAME+"/"+self.PREDICT_NAME
      predict_pb = fileOp.readFile(filePath = predict_path ,openMode = "rb")

      workspace.ResetWorkspace()
      #2.use the string to generate an instance of Net
      init_net = caffe2_pb2.NetDef()
      init_net.ParseFromString(init_pb)
      init_net.name = unicode("test_init")
      #print("wb->",workspace.StringifyNetName("test_init"))
      #init_net = Net(init_net)
      predict_net = caffe2_pb2.NetDef()
      predict_net.ParseFromString(predict_pb)
      #print(predict_net)

      predict_net.name = unicode("test")
      workspace.RunNetOnce(init_net)
      workspace.FeedBlob(unicode("data"),image)

      workspace.CreateNet(predict_net)
      workspace.RunNet(predict_net.name)
      output = predict_net.external_output
      #print("run predict and fetch output ",workspace.FetchBlob(output[0]))
      return workspace.FetchBlob(output[0])

分析：be patient and be smart

   def testImage(self,image_location,image_size):

#1. 读取核心文件的内容：图片文件、init_net 和 predict_net；后两个文件读出的是protobuf 格式的string，需要用特定的_pb2.py文件进行解析。
      mean = self.readMeanFile()
      image = ImageProcess(image_location,image_size,mean).getImage()#读图片内容
      fileOp = FileOperation()
      init_path = self.MODEL_PATH+"/"+self.MODEL_NAME+"/"+self.INIT_NAME
      init_pb = fileOp.readFile(filePath = init_path ,openMode = "rb")#读init

      predict_path = self.MODEL_PATH+"/"+self.MODEL_NAME+"/"+self.PREDICT_NAME
      predict_pb = fileOp.readFile(filePath = predict_path ,openMode = "rb")#读predict

#2. workspace中存放了大量的blobs，每个blob都是{key，value}的形式：key是blob的名称，value是一个叫做tensor的数据结构，类似于python中的array。

2.1 worspace的作用

Similar to Matlab, the Caffe2 workspace consists of blobs you create and store in memory. A blob is actually a typed pointer that can store any type of C++ objects, but Tensor is the most common type stored in a blob. 

https://github.com/caffe2/tutorials/blob/master/Basics.ipynb

https://caffe2.ai/docs/tutorial-basics-of-caffe2.html

     workspace.ResetWorkspace()#置空当前的workspace。
      #2.use the string to generate an instance of Net
      init_net = caffe2_pb2.NetDef()
      init_net.ParseFromString(init_pb)
      init_net.name = unicode("test_init") #创建pb的过程中缺失了filed name,这里需要补充上。如果不明确给出name，将pb转为ND对象时，ND会自动对name赋值。

      predict_net = caffe2_pb2.NetDef()
      predict_net.ParseFromString(predict_pb)

      predict_net.name = unicode("test")

#3.真正的运行net：Then there are two ways to run a net from Python. We will do the first option in the example below.

1. Call workspace.RunNetOnce(), which instantiates, runs and immediately destructs the network。
2. Call workspace.CreateNet() to create the C++ net object owned by the workspace, then call workspace.RunNet(), passing the name of the network to it

      workspace.RunNetOnce(init_net) #所有的input和output内容都实例化成一个blob，存放到了workspace里。但是，data字段没有填入内容。
      workspace.FeedBlob(unicode("data"),image)#把image信息填入到data字段

      workspace.CreateNet(predict_net)#实例化对象，创建对应的blob，并挂载到workspace上。
      workspace.RunNet(predict_net.name)#运行net
      output = predict_net.external_output
      #print("run predict and fetch output ",workspace.FetchBlob(output[0]))#取对应的结果
      return workspace.FetchBlob(output[0])