本文共 2680 字，大约阅读时间需要 8 分钟。

1. caffe介绍

caffe是贾扬清大神开发的一套系统，caffe2是重构后的版本。其基本结构为：

• 使用blob存储数据
• 使用operator连接blob数据，定义了计算规则
• 使用net组合operator形成批次计算的层
• 使用workspace管理网络、计算符和数据

2. 基本操作

2.1 数据

import numpy as npimport timefrom caffe2.python import core, workspacefrom caffe2.proto import caffe2_pb2X = np.random.randn(2, 3).astype(np.float32)print("Generated X from numpy:\n{}".format(X))workspace.FeedBlob("X", X)print(workspace.Blobs())print(workspace.FetchBlob('X'))

Blobs命令得到数据名列表

2.2 操作符

op = core.CreateOperator(    "Relu", # The type of operator that we want to run    ["X"], # A list of input blobs by their names    ["Y"], # A list of output blobs by their names    'active1')workspace.RunOperatorOnce(op)workspace.FetchBlobs('Y')op2 = core.CreateOperator(    "GaussianFill",    [], # GaussianFill does not need any parameters.    ["W"],    shape=[100, 100], # shape argument as a list of ints.    mean=1.0,  # mean as a single float    std=1.0, # std as a single float)workspace.RunOperatorOnce(op2)temp = workspace.FetchBlob("Z")

使用workspace来控制运行。

2.3 网络

net = core.Net("linear_net")X = net.GaussianFill([], ["X"], mean=0.0, std=1.0, shape=[2, 3], run_once=0)W = net.GaussianFill([], ["W"], mean=0.0, std=1.0, shape=[5, 3], run_once=0)b = net.ConstantFill([], ["b"], shape=[5,], value=1.0, run_once=0)Y = net.FC([X,W,b],['Y'])print(net.Proto())workspace.RunNetOnce(net)for name in workspace.Blobs():    print("{}:\n{}".format(name, workspace.FetchBlob(name)))

在python中有两种方法来运行一个net：

from caffe2.python import net_drawerfrom IPython import displaygraph = net_drawer.GetPydotGraph(net, rankdir="LR")display.Image(graph.create_png(), width=400)

2.4 运行推理

with open(INIT_NET) as f:    init_net = f.read()with open(PREDICT_NET) as f:    predict_net = f.read()p = workspace.Predictor(init_net, predict_net)results = p.run([img])

2.5 brew包

caffe2为了更方便调用和管理，可以通过导入brew这个包来调用帮助函数。像fc层的实现就可以使用：

from caffe2.python import brewbrew.fc(model, blob_in, blob_out, ...)

我们使用brew构造网络就十分简单，下面的代码就构造了一个LeNet模型：

from caffe2.python import brewdef AddLeNetModel(model, data):    conv1 = brew.conv(model, data, 'conv1', 1, 20, 5)    pool1 = brew.max_pool(model, conv1, 'pool1', kernel=2, stride=2)    conv2 = brew.conv(model, pool1, 'conv2', 20, 50, 5)    pool2 = brew.max_pool(model, conv2, 'pool2', kernel=2, stride=2)    fc3 = brew.fc(model, pool2, 'fc3', 50 * 4 * 4, 500)    fc3 = brew.relu(model, fc3, fc3)    pred = brew.fc(model, fc3, 'pred', 500, 10)    softmax = brew.softmax(model, pred, 'softmax')

caffe2 使用brew提供很多构造网络的帮助函数，大大简化了我们构建网络的过程。但实际上，这些只是封装的结果，网络构造的原理和之前说的使用operators构建的原理是一样的。