caffe训练自己的手写数字

最新推荐文章于 2021-11-19 19:46:22 发布

mjiansun

最新推荐文章于 2021-11-19 19:46:22 发布

阅读量2.3k

点赞数 2

分类专栏： Caffe

Caffe 专栏收录该内容

58 篇文章 5 订阅

订阅专栏

转自：http://blog.csdn.net/houwenbin1986/article/details/52956101

搭建好caffe Python环境后，我们都需要跑通mnist和imagenet示例，感谢博主：http://www.cnblogs.com/denny402/p/5684431.html

官网提供的mnist数据并不是图片，但我们以后做的实际项目可能是图片。因此有些人并不知道该怎么办。在此我将mnist数据进行了转化，变成了一张张的图片，我们练习就从图片开始。mnist图片数据我放在了百度云盘。

mnist图片数据下载：http://pan.baidu.com/s/1pLMV4Kz

数据分成了训练集（60000张共10类）和测试集（共10000张10类），每个类别放在一个单独的文件夹里。并且将所有的图片，都生成了txt列表清单（train.txt和test.txt)。大家下载下来后，直接解压到当前用户根目录下就可以了。由于我是在windows下压缩的，因此是winrar文件。如果大家要在linux下解压缩，需要安装rar的linux版本，也是十分简单

记录一下自己的实验步骤：

(root) [root@localhost lenet5]# python verify.py 
WARNING: Logging before InitGoogleLogging() is written to STDERR
I1028 14:14:45.532413  3195 net.cpp:49] Initializing net from parameters: 
name: "Lenet"
input: "data"
input_dim: 1
input_dim: 3
input_dim: 28
input_dim: 28
state {
  phase: TEST
}
layer {
  name: "Convolution1"
  type: "Convolution"
  bottom: "data"
  top: "Convolution1"
  convolution_param {
    num_output: 20
    pad: 0
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "Pooling1"
  type: "Pooling"
  bottom: "Convolution1"
  top: "Pooling1"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "Convolution2"
  type: "Convolution"
  bottom: "Pooling1"
  top: "Convolution2"
  convolution_param {
    num_output: 50
    pad: 0
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "Pooling2"
  type: "Pooling"
  bottom: "Convolution2"
  top: "Pooling2"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "InnerProduct1"
  type: "InnerProduct"
  bottom: "Pooling2"
  top: "InnerProduct1"
  inner_product_param {
    num_output: 500
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "ReLU1"
  type: "ReLU"
  bottom: "InnerProduct1"
  top: "InnerProduct1"
}
layer {
  name: "InnerProduct2"
  type: "InnerProduct"
  bottom: "InnerProduct1"
  top: "InnerProduct2"
  inner_product_param {
    num_output: 10
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "Softmax1"
  type: "Softmax"
  bottom: "InnerProduct2"
  top: "Softmax1"
}
I1028 14:14:45.532510  3195 net.cpp:413] Input 0 -> data
I1028 14:14:45.532536  3195 layer_factory.hpp:77] Creating layer Convolution1
I1028 14:14:45.532551  3195 net.cpp:106] Creating Layer Convolution1
I1028 14:14:45.532555  3195 net.cpp:454] Convolution1 <- data
I1028 14:14:45.532562  3195 net.cpp:411] Convolution1 -> Convolution1
I1028 14:14:45.533006  3195 net.cpp:150] Setting up Convolution1
I1028 14:14:45.533018  3195 net.cpp:157] Top shape: 1 20 24 24 (11520)
I1028 14:14:45.533022  3195 net.cpp:165] Memory required for data: 46080
I1028 14:14:45.533035  3195 layer_factory.hpp:77] Creating layer Pooling1
I1028 14:14:45.533044  3195 net.cpp:106] Creating Layer Pooling1
I1028 14:14:45.533048  3195 net.cpp:454] Pooling1 <- Convolution1
I1028 14:14:45.533054  3195 net.cpp:411] Pooling1 -> Pooling1
I1028 14:14:45.533067  3195 net.cpp:150] Setting up Pooling1
I1028 14:14:45.533073  3195 net.cpp:157] Top shape: 1 20 12 12 (2880)
I1028 14:14:45.533077  3195 net.cpp:165] Memory required for data: 57600
I1028 14:14:45.533082  3195 layer_factory.hpp:77] Creating layer Convolution2
I1028 14:14:45.533088  3195 net.cpp:106] Creating Layer Convolution2
I1028 14:14:45.533093  3195 net.cpp:454] Convolution2 <- Pooling1
I1028 14:14:45.533099  3195 net.cpp:411] Convolution2 -> Convolution2
I1028 14:14:45.533278  3195 net.cpp:150] Setting up Convolution2
I1028 14:14:45.533285  3195 net.cpp:157] Top shape: 1 50 8 8 (3200)
I1028 14:14:45.533290  3195 net.cpp:165] Memory required for data: 70400
I1028 14:14:45.533298  3195 layer_factory.hpp:77] Creating layer Pooling2
I1028 14:14:45.533304  3195 net.cpp:106] Creating Layer Pooling2
I1028 14:14:45.533308  3195 net.cpp:454] Pooling2 <- Convolution2
I1028 14:14:45.533314  3195 net.cpp:411] Pooling2 -> Pooling2
I1028 14:14:45.533323  3195 net.cpp:150] Setting up Pooling2
I1028 14:14:45.533329  3195 net.cpp:157] Top shape: 1 50 4 4 (800)
I1028 14:14:45.533332  3195 net.cpp:165] Memory required for data: 73600
I1028 14:14:45.533336  3195 layer_factory.hpp:77] Creating layer InnerProduct1
I1028 14:14:45.533345  3195 net.cpp:106] Creating Layer InnerProduct1
I1028 14:14:45.533349  3195 net.cpp:454] InnerProduct1 <- Pooling2
I1028 14:14:45.533355  3195 net.cpp:411] InnerProduct1 -> InnerProduct1
I1028 14:14:45.536108  3195 net.cpp:150] Setting up InnerProduct1
I1028 14:14:45.536118  3195 net.cpp:157] Top shape: 1 500 (500)
I1028 14:14:45.536121  3195 net.cpp:165] Memory required for data: 75600
I1028 14:14:45.536130  3195 layer_factory.hpp:77] Creating layer ReLU1
I1028 14:14:45.536136  3195 net.cpp:106] Creating Layer ReLU1
I1028 14:14:45.536140  3195 net.cpp:454] ReLU1 <- InnerProduct1
I1028 14:14:45.536146  3195 net.cpp:397] ReLU1 -> InnerProduct1 (in-place)
I1028 14:14:45.536154  3195 net.cpp:150] Setting up ReLU1
I1028 14:14:45.536159  3195 net.cpp:157] Top shape: 1 500 (500)
I1028 14:14:45.536161  3195 net.cpp:165] Memory required for data: 77600
I1028 14:14:45.536165  3195 layer_factory.hpp:77] Creating layer InnerProduct2
I1028 14:14:45.536171  3195 net.cpp:106] Creating Layer InnerProduct2
I1028 14:14:45.536175  3195 net.cpp:454] InnerProduct2 <- InnerProduct1
I1028 14:14:45.536181  3195 net.cpp:411] InnerProduct2 -> InnerProduct2
I1028 14:14:45.536227  3195 net.cpp:150] Setting up InnerProduct2
I1028 14:14:45.536233  3195 net.cpp:157] Top shape: 1 10 (10)
I1028 14:14:45.536237  3195 net.cpp:165] Memory required for data: 77640
I1028 14:14:45.536243  3195 layer_factory.hpp:77] Creating layer Softmax1
I1028 14:14:45.536250  3195 net.cpp:106] Creating Layer Softmax1
I1028 14:14:45.536254  3195 net.cpp:454] Softmax1 <- InnerProduct2
I1028 14:14:45.536259  3195 net.cpp:411] Softmax1 -> Softmax1
I1028 14:14:45.536270  3195 net.cpp:150] Setting up Softmax1
I1028 14:14:45.536275  3195 net.cpp:157] Top shape: 1 10 (10)
I1028 14:14:45.536279  3195 net.cpp:165] Memory required for data: 77680
I1028 14:14:45.536283  3195 net.cpp:228] Softmax1 does not need backward computation.
I1028 14:14:45.536288  3195 net.cpp:228] InnerProduct2 does not need backward computation.
I1028 14:14:45.536293  3195 net.cpp:228] ReLU1 does not need backward computation.
I1028 14:14:45.536296  3195 net.cpp:228] InnerProduct1 does not need backward computation.
I1028 14:14:45.536300  3195 net.cpp:228] Pooling2 does not need backward computation.
I1028 14:14:45.536304  3195 net.cpp:228] Convolution2 does not need backward computation.
I1028 14:14:45.536309  3195 net.cpp:228] Pooling1 does not need backward computation.
I1028 14:14:45.536314  3195 net.cpp:228] Convolution1 does not need backward computation.
I1028 14:14:45.536317  3195 net.cpp:270] This network produces output Softmax1
I1028 14:14:45.536325  3195 net.cpp:283] Network initialization done.
I1028 14:14:45.539165  3195 net.cpp:816] Ignoring source layer ImageData1
I1028 14:14:45.539448  3195 net.cpp:816] Ignoring source layer SoftmaxWithLoss1
[ 0.  0.  0.  0.  0.  0.  0.  0.  0.  1.]
the class is: 9
(root) [root@localhost lenet5]#

1、生成必须的训练、测试、网络配置文件，脚本 1.run_lenet5.py

# -*- coding: utf-8 -*-

import caffe
from caffe import layers as L,params as P,proto,to_proto

#设定文件的保存路径
root = '/root/AI/lenet5/'                           #根目录
train_list = root+'mnist/train/train.txt'     #训练图片列表
test_list = root+'mnist/test/test.txt'        #测试图片列表
train_proto = root+'mnist/train.prototxt'     #训练配置文件
test_proto = root+'mnist/test.prototxt'       #测试配置文件
solver_proto = root+'mnist/solver.prototxt'   #参数文件

#编写一个函数，生成配置文件prototxt
def Lenet(img_list,batch_size,include_acc=False):
    #第一层，数据输入层，以ImageData格式输入
    data, label = L.ImageData(source=img_list, batch_size=batch_size, ntop=2,root_folder=root,
        transform_param=dict(scale= 0.00390625))
    #第二层：卷积层
    conv1=L.Convolution(data, kernel_size=5, stride=1,num_output=20, pad=0,weight_filler=dict(type='xavier'))
    #池化层
    pool1=L.Pooling(conv1, pool=P.Pooling.MAX, kernel_size=2, stride=2)
    #卷积层
    conv2=L.Convolution(pool1, kernel_size=5, stride=1,num_output=50, pad=0,weight_filler=dict(type='xavier'))
    #池化层
    pool2=L.Pooling(conv2, pool=P.Pooling.MAX, kernel_size=2, stride=2)
    #全连接层
    fc3=L.InnerProduct(pool2, num_output=500,weight_filler=dict(type='xavier'))
    #激活函数层
    relu3=L.ReLU(fc3, in_place=True)
    #全连接层
    fc4 = L.InnerProduct(relu3, num_output=10,weight_filler=dict(type='xavier'))
    #softmax层
    loss = L.SoftmaxWithLoss(fc4, label)
    
    if include_acc:             # test阶段需要有accuracy层
        acc = L.Accuracy(fc4, label)
        return to_proto(loss, acc)
    else:
        return to_proto(loss)
    
def write_net():
    #写入train.prototxt
    with open(train_proto, 'w') as f:
        f.write(str(Lenet(train_list,batch_size=64)))

    #写入test.prototxt    
    with open(test_proto, 'w') as f:
        f.write(str(Lenet(test_list,batch_size=100, include_acc=True)))

#编写一个函数，生成参数文件
def gen_solver(solver_file,train_net,test_net):
    s = proto.caffe_pb2.SolverParameter()
    s.train_net =train_net
    s.test_net.append(test_net)
    s.test_interval = 938    #60000/64，测试间隔参数：训练完一次所有的图片，进行一次测试  
    s.test_iter.append(100)  #10000/100 测试迭代次数，需要迭代100次，才完成一次所有数据的测试
    s.max_iter = 9380       #10 epochs , 938*10，最大训练次数
    s.base_lr = 0.01    #基础学习率
    s.momentum = 0.9    #动量
    s.weight_decay = 5e-4  #权值衰减项
    s.lr_policy = 'step'   #学习率变化规则
    s.stepsize=3000         #学习率变化频率
    s.gamma = 0.1          #学习率变化指数
    s.display = 20         #屏幕显示间隔
    s.snapshot = 938       #保存caffemodel的间隔
    s.snapshot_prefix = root + 'mnist/lenet'   #caffemodel前缀
    s.type ='SGD'         #优化算法
    s.solver_mode = proto.caffe_pb2.SolverParameter.CPU    #加速
    #写入solver.prototxt
    with open(solver_file, 'w') as f:
        f.write(str(s))

#开始训练
def training(solver_proto):
    #caffe.set_device(0)
    #caffe.set_mode_gpu()
    caffe.set_mode_cpu()
    solver = caffe.SGDSolver(solver_proto)
    solver.solve()

if __name__ == '__main__':
    write_net()
    gen_solver(solver_proto,train_proto,test_proto) 
    training(solver_proto)

2、训练好模型后，生成识别用的网络模型，脚本 2.mkdeploy.py

# -*- coding: utf-8 -*-

import caffe
from caffe import layers as L,params as P,to_proto

root = 'D:/MyWorks/caffe-windows-master/examples/lenet5/'
deploy = root+'mnist/deploy.prototxt'    #文件保存路径

def create_deploy():
    #少了第一层，data层
    conv1 = L.Convolution(bottom='data', kernel_size=5, stride=1,num_output=20, pad=0,weight_filler=dict(type='xavier'))
    pool1 = L.Pooling(conv1, pool=P.Pooling.MAX, kernel_size=2, stride=2)
    conv2 = L.Convolution(pool1, kernel_size=5, stride=1,num_output=50, pad=0,weight_filler=dict(type='xavier'))
    pool2 = L.Pooling(conv2, pool=P.Pooling.MAX, kernel_size=2, stride=2)
    fc3 = L.InnerProduct(pool2, num_output=500,weight_filler=dict(type='xavier'))
    relu3 = L.ReLU(fc3, in_place=True)
    fc4 = L.InnerProduct(relu3, num_output=10,weight_filler=dict(type='xavier'))
    #最后没有accuracy层，但有一个Softmax层
    prob = L.Softmax(fc4)
    return to_proto(prob)

def write_deploy(): 
    with open(deploy, 'w') as f:
        f.write('name:"Lenet"\n')
        f.write('input:"data"\n')
        f.write('input_dim:1\n')
        f.write('input_dim:3\n')
        f.write('input_dim:28\n')
        f.write('input_dim:28\n')
        f.write(str(create_deploy()))

if __name__ == '__main__':
    write_deploy()

3、最后是读取待识别的图片进行数字识别，脚本 3.verify.py

#coding=utf-8

import caffe
import numpy as np

root = 'D:/MyWorks/caffe-windows-master/examples/lenet5/'   #根目录
deploy = root + 'mnist/deploy.prototxt'    #deploy文件
caffe_model = root + 'mnist/lenet_iter_9380.caffemodel'   #训练好的 caffemodel
img = root + 'mnist/test/9/00479.png'    #随机找的一张待测图片
labels_filename = root + 'mnist/test/labels.txt'  #类别名称文件，将数字标签转换回类别名称

net = caffe.Net(deploy,caffe_model,caffe.TEST)   #加载model和network

#图片预处理设置
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})  #设定图片的shape格式(1,3,28,28)
transformer.set_transpose('data', (2,0,1))    #改变维度的顺序，由原始图片(28,28,3)变为(3,28,28)
#transformer.set_mean('data', np.load(mean_file).mean(1).mean(1))    #减去均值，前面训练模型时没有减均值，这儿就不用
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

im = caffe.io.load_image(img)                   #加载图片
net.blobs['data'].data[...] = transformer.preprocess('data',im)      #执行上面设置的图片预处理操作，并将图片载入到blob中

#执行测试
out = net.forward()

labels = np.loadtxt(labels_filename, str, delimiter='\t')   #读取类别名称文件
prob= net.blobs['Softmax1'].data[0].flatten() #取出最后一层（Softmax）属于某个类别的概率值，并打印
print prob
order=prob.argsort()[-1]  #将概率值排序，取出最大值所在的序号 
print 'the class is:',labels[order]   #将该序号转换成对应的类别名称，并打印

mjiansun

关注

2
点赞
踩
6

收藏

觉得还不错? 一键收藏
0
评论
caffe训练自己的手写数字

转自：http://blog.csdn.net/houwenbin1986/article/details/52956101搭建好caffe Python环境后，我们都需要跑通mnist和imagenet示例，感谢博主：http://www.cnblogs.com/denny402/p/5684431.html官网提供的mnist数据并不是图片，但我们以后做的实际项目可能是图片
复制链接

扫一扫

专栏目录