本文共 4230 字，大约阅读时间需要 14 分钟。

一、准备样本数据

获取minist的数据包。这个版本是四个数据包。

learning@learning-virtual-machine:~/caffe/data/mnist$ ./get_mnist.sh

下载过程如下：

Downloading...
--2016-05-11 17:35:10--  http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz
Resolving yann.lecun.com (yann.lecun.com)... 128.122.47.89
Connecting to yann.lecun.com (yann.lecun.com)|128.122.47.89|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 9912422 (9.5M) [application/x-gzip]
Saving to: ‘train-images-idx3-ubyte.gz’
train-images-idx 100%[===========>]   9.45M   225KB/s   in 73s    2016-05-11 17:36:24 (133 KB/s) - ‘train-images-idx3-ubyte.gz’ saved [9912422/9912422]

--2016-05-11 17:36:38--  http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz
Resolving yann.lecun.com (yann.lecun.com)... 128.122.47.89
Connecting to yann.lecun.com (yann.lecun.com)|128.122.47.89|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 28881 (28K) [application/x-gzip]
Saving to: ‘train-labels-idx1-ubyte.gz’
train-labels-idx 100%[===========>]  28.20K  1.41KB/s   in 7.5s   2016-05-11 17:36:49 (3.75 KB/s) - ‘train-labels-idx1-ubyte.gz’ saved [28881/28881]

--2016-05-11 17:36:49--  http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz
Resolving yann.lecun.com (yann.lecun.com)... 128.122.47.89
Connecting to yann.lecun.com (yann.lecun.com)|128.122.47.89|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 1648877 (1.6M) [application/x-gzip]
Saving to: ‘t10k-images-idx3-ubyte.gz’
t10k-images-idx3 100%[===========>]   1.57M  71.9KB/s   in 19s    2016-05-11 17:37:08 (85.5 KB/s) - ‘t10k-images-idx3-ubyte.gz’ saved [1648877/1648877]

--2016-05-11 17:37:09--  http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz
Resolving yann.lecun.com (yann.lecun.com)... 128.122.47.89
Connecting to yann.lecun.com (yann.lecun.com)|128.122.47.89|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 4542 (4.4K) [application/x-gzip]
Saving to: ‘t10k-labels-idx1-ubyte.gz’
t10k-labels-idx1 100%[===========>]   4.44K  --.-KB/s   in 0s     2016-05-11 17:37:09 (31.5 MB/s) - ‘t10k-labels-idx1-ubyte.gz’ saved [4542/4542]

get_mnist.sh 代码

#!/usr/bin/env sh
# This scripts downloads the mnist data and unzips it.
DIR="$( cd "$(dirname "$0")" ; pwd -P)"
cd $DIR
echo "Downloading..."
for fname in train-images-idx3-ubyte train-labels-idx1-ubyte t10k-images-idx3-ubyte t10k-labels-idx1-ubyte
do
    if [ ! -e $fname ]; then
        wget --no-check-certificate http://yann.lecun.com/exdb/mnist/${fname}.gz
        gunzip ${fname}.gz
    fi
done

二、训练

learning@learning-virtual-machine:~/caffe$ ./examples/mnist/train_lenet.sh

出现问题：

I0511 17:52:25.115056 63914 caffe.cpp:185] Using GPUs 0
F0511 17:52:25.116345 63914 common.cpp:66] Cannot use GPU in CPU-only Caffe: check mode.

解决方案：

learning@learning-virtual-machine:~/caffe/examples/mnist$ sudo gedit lenet_solver.prototxt

在文件中添加以下内容：

# 检查是否有GPU可用
device: GPU

2. 运行训练脚本

learning@learning-virtual-machine:~/caffe/examples/mnist$ ./train_lenet.sh

如果问题依然存在，请检查以下内容：

• CUDA 驱动是否正确安装
• CUDA 库文件是否存在
• 重新启动系统

三、测试

learning@learning-virtual-machine:~/caffe$ ./build/tools/caffe.bin test -model=examples/mnist/lenet_train_test.prototxt -weights=examples/mnist/lenet_iter_10000.caffemodel

I0512 09:18:41.455747  3503 caffe.cpp:275] Batch 44, loss = 0.0137619
I0512 09:18:41.671058  3503 caffe.cpp:275] Batch 45, accuracy = 0.99
I0512 09:18:41.671362  3503 caffe.cpp:275] Batch 45, loss = 0.0446652
I0512 09:18:41.910468  3503 caffe.cpp:275] Batch 46, accuracy = 1
I0512 09:18:41.910781  3503 caffe.cpp:275] Batch 46, loss = 0.00462838
I0512 09:18:42:082020  3503 caffe.cpp:275] Batch 47, accuracy = 0.99
I0512 09:18:42:082260  3503 caffe.cpp:275] Batch 47, loss = 0.0215265
I0512 09:18:42:297307  3503 caffe.cpp:275] Batch 48, accuracy = 0.96
I0512 09:18:42:301200  3503 caffe.cpp:275] Batch 48, loss = 0.0964929
I0512 09:18:42:576354  3503 caffe.cpp:275] Batch 49, accuracy = 1
I0512 09:18:42:576627  3503 caffe.cpp:275] Batch 49, loss = 0.00345927
I0512 09:18:42:576843  3503 caffe.cpp:280] Loss: 0.0427004
I0512 09:18:42:576843  3503 caffe.cpp:292] accuracy = 0.9872
I0512 09:18:42:576954  3503 caffe.cpp:292] loss = 0.0427004 (* 1 = 0.0427004 loss)

参考资料

• Caffe GPU 支持