Caffe mnist数据集训练

文章作者：Tyan
博客：noahsnail.com  |  CSDN  |  简书

本文是使用基于docker的caffe进行实验。docker启动过程请看docker相关的博客。训练过程如下：

# 进入caffe目录
$:/workspace# cd /opt/caffe/

# 下载mnist数据集
$:/opt/caffe# ./data/mnist/get_mnist.sh 

# 将mnist数据集转换成lmdb
$:/opt/caffe# sh examples/mnist/create_mnist.sh

# 修改examples/mnist/lenet_solver.prototxt
将GPU改为CPU

# 运行examples/mnist/train_lenet.sh
$:/opt/caffe# sh examples/mnist/train_lenet.sh


# 训练结果
I1205 09:52:22.457670    33 solver.cpp:329] Iteration 10000, loss = 0.0035313
I1205 09:52:22.457761    33 solver.cpp:349] Iteration 10000, Testing net (#0)
I1205 09:52:27.329227    33 solver.cpp:416]     Test net output #0: accuracy = 0.9913
I1205 09:52:27.329305    33 solver.cpp:416]     Test net output #1: loss = 0.026992 (* 1 = 0.026992 loss)
I1205 09:52:27.329318    33 solver.cpp:334] Optimization Done.
I1205 09:52:27.329326    33 caffe.cpp:254] Optimization Done.