caffe 利用cifar10进行训练

下载cifar10数据集

​​ http://www.cs.toronto.edu/~kriz/cifar.html​​

生成lmdb文件

import os
import shutil

EXAMPLE="./cifar10"
DATA="./cifar10"
DBTYPE="lmdb"

print "Creating "+DBTYPE+"..."

path1=EXAMPLE+"\\cifar10_train_"+DBTYPE
path2=EXAMPLE+"\\cifar10_test_"+DBTYPE

if os.path.exists(path1):
    shutil.rmtree(path1)
if os.path.exists(path2):
    shutil.rmtree(path2)

caffe_build="D:\\software\\caffe\\bin"
cmd1=caffe_build+"\\convert_cifar_data.exe "+DATA+" "+EXAMPLE+" "+DBTYPE
print cmd1
os.system(cmd1)

print "Computing image mean..."

cmd2=caffe_build+"\\compute_image_mean.exe --backend="+DBTYPE+" "+EXAMPLE+"\\cifar10_train_"+DBTYPE+" "+EXAMPLE+"\\mean.binaryproto"
print cmd2
os.system(cmd2)

使用compute_image_mean.exe 、convert_cifar_data.exe

利用caffe生成net、solver文件并训练

#coding='utf-8'
import lmdb
import caffe
from caffe.proto import caffe_pb2
from caffe import layers as L
from caffe import params as P
from matplotlib import pyplot as plt
import numpy as np


data_path = "cifar10/"
train_net_file = "auto_train.prototxt"
test_net_file = "auto_test.prototxt"
solver_file = "auto_slover.prototxt"


def net(datafile,mean_file,batch_size):
  n = caffe.NetSpec()
  n.data,n.label=L.Data(source=datafile, backend = P.Data.LMDB, batch_size=batch_size, ntop=2, transform_param=dict(scale=1.0/255.0, mean_file=mean_file))
  n.ip1 = L.InnerProduct(n.data, num_output=200, weight_filler=dict(type='xavier'))
  n.relu1 = L.ReLU(n.ip1, in_place=True)
  n.ip2 = L.InnerProduct(n.relu1, num_output=10, weight_filler=dict(type='xavier'))
  n.loss = L.SoftmaxWithLoss(n.ip2, n.label)
  n.accu = L.Accuracy(n.ip2, n.label, include={'phase':caffe.TEST})
  return n.to_proto()

def config():
  s = caffe_pb2.SolverParameter()
  s.train_net = train_net_file
  s.test_net.append(test_net_file)
  s.test_interval = 500  
  s.test_iter.append(100) 
  s.display = 500
  s.max_iter = 10000   
  s.weight_decay = 0.005
  s.base_lr = 0.1
  s.lr_policy = "step"
  s.gamma = 0.1
  s.stepsize = 5000
  s.solver_mode = caffe_pb2.SolverParameter.GPU

  with open(solver_file, 'w') as f:
    f.write(str(s))



def main():
  with open( train_net_file, 'w') as f:
    f.write(str(net(data_path+'cifar10_train_lmdb',  data_path+'mean.binaryproto', 200)))
  with open( test_net_file, 'w') as f:
    f.write(str(net(data_path+'cifar10_test_lmdb',  data_path+'mean.binaryproto', 100)))
  config()
  solver = caffe.get_solver(solver_file)


  niter = 2001
  train_loss = np.zeros(niter)
  test_acc = np.zeros(niter)

  for it in range(niter):
    solver.step(1)  
    train_loss[it] = solver.net.blobs['loss'].data
    test_acc[it] = solver.test_nets[0].blobs['accu'].data


  _, ax1 = plt.subplots()
  ax2 = ax1.twinx()
  ax1.plot(np.arange(niter), train_loss)
  ax2.plot(np.arange(niter), test_acc, 'r')
  ax1.set_xlabel('iteration')
  ax1.set_ylabel('train loss')
  ax2.set_ylabel('test accuracy')
  _.savefig('result.png')


main()