裝好caffe之后，下面我們來跑它自帶的第一個例子，在mnist進行實驗，參見http://caffe.berkeleyvision.org/gathered/examples/mnist.html
（1）caffe在mnist自帶的是使用leNet的網絡結構。
1.簡介：
LeNet論文是Yan LeCun在1989年的論文Gradient-Based Learning Applied to Document Recognition
http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf,這是CNN第一篇經典之作.
minst是手寫字體庫，CNN的LeNet的廣泛應用於美國的支票手寫字體識別。
2.LeNet網絡結構
（2）在mnist 進行實驗
均在caffe根目錄下執行
- 1.下載數據： ./data/mnist/get_mnist.sh,
- 2.生成lmdb文件：./example/mnist/create_minist.sh
- 此時在當前目錄下生成 mnist_train_lmdb,mnist_test_lmdb文件
- 3.配置網絡 ：letNet網絡的定義在。./examples/mnist/lenet_train_test.prototxt文件中
lenet_train_test.prototxt內容如下
name: "LeNet"
2 layers {
3 name: "mnist"
4 type: DATA
5 top: "data"
6 top: "label"
7 data_param {
8 source: "examples/mnist/mnist_train_lmdb"
9 backend: LMDB
10 batch_size: 64
11 }
12 transform_param {
13 scale: 0.00390625
14 }
15 include: { phase: TRAIN }
16 }
17 layers {
18 name: "mnist"
19 type: DATA
20 top: "data"
21 top: "label"
22 data_param {
23 source: "examples/mnist/mnist_test_lmdb"
24 backend: LMDB
25 batch_size: 100
26 }
27 transform_param {
28 scale: 0.00390625
29 }
30 include: { phase: TEST }
31 }
32
33 layers {
34 name: "conv1"
35 type: CONVOLUTION
36 bottom: "data"
37 top: "conv1"
38 blobs_lr: 1
39 blobs_lr: 2
40 convolution_param {
41 num_output: 20
42 kernel_size: 5
43 stride: 1
44 weight_filler {
45 type: "xavier"
46 }
47 bias_filler {
48 type: "constant"
49 }
50 }
51 }
52 layers {
53 name: "pool1"
54 type: POOLING
55 bottom: "conv1"
56 top: "pool1"
57 pooling_param {
58 pool: MAX
59 kernel_size: 2
60 stride: 2
61 }
62 }
63 layers {
64 name: "conv2"
65 type: CONVOLUTION
66 bottom: "pool1"
67 top: "conv2"
68 blobs_lr: 1
69 blobs_lr: 2
70 convolution_param {
71 num_output: 50
72 kernel_size: 5
73 stride: 1
74 weight_filler {
75 type: "xavier"
76 }
77 bias_filler {
78 type: "constant"
79 }
80 }
81 }
82 layers {
83 name: "pool2"
84 type: POOLING
85 bottom: "conv2"
86 top: "pool2"
87 pooling_param {
88 pool: MAX
89 kernel_size: 2
90 stride: 2
91 }
92 }
93 layers {
94 name: "ip1"
95 type: INNER_PRODUCT
96 bottom: "pool2"
97 top: "ip1"
98 blobs_lr: 1
99 blobs_lr: 2
100 inner_product_param {
101 num_output: 500
102 weight_filler {
103 type: "xavier"
104 }
105 bias_filler {
106 type: "constant"
107 }
108 }
109 }
110 layers {
111 name: "relu1"
112 type: RELU
113 bottom: "ip1"
114 top: "ip1"
115 }
116 layers {
117 name: "ip2"
118 type: INNER_PRODUCT
119 bottom: "ip1"
120 top: "ip2"
121 blobs_lr: 1
122 blobs_lr: 2
123 inner_product_param {
124 num_output: 10
125 weight_filler {
126 type: "xavier"
127 }
128 bias_filler {
129 type: "constant"
130 }
131 }
132 }
133 layers {
134 name: "accuracy"
135 type: ACCURACY
136 bottom: "ip2"
137 bottom: "label"
138 top: "accuracy"
139 include: { phase: TEST }
140 }
141 layers {
142 name: "loss"
143 type: SOFTMAX_LOSS
144 bottom: "ip2"
145 bottom: "label"
146 top: "loss"
147 }
name: "LeNet"
2 layers {
3 name: "mnist"
4 type: DATA
5 top: "data"
6 top: "label"
7 data_param {
8 source: "examples/mnist/mnist_train_lmdb"
9 backend: LMDB
10 batch_size: 64
11 }
12 transform_param {
13 scale: 0.00390625
14 }
15 include: { phase: TRAIN }
16 }
17 layers {
18 name: "mnist"
19 type: DATA
20 top: "data"
21 top: "label"
22 data_param {
23 source: "examples/mnist/mnist_test_lmdb"
24 backend: LMDB
25 batch_size: 100
26 }
27 transform_param {
28 scale: 0.00390625
29 }
30 include: { phase: TEST }
31 }
32
33 layers {
34 name: "conv1"
35 type: CONVOLUTION
36 bottom: "data"
37 top: "conv1"
38 blobs_lr: 1
39 blobs_lr: 2
40 convolution_param {
41 num_output: 20
42 kernel_size: 5
43 stride: 1
44 weight_filler {
45 type: "xavier"
46 }
47 bias_filler {
48 type: "constant"
49 }
50 }
51 }
52 layers {
53 name: "pool1"
54 type: POOLING
55 bottom: "conv1"
56 top: "pool1"
57 pooling_param {
58 pool: MAX
59 kernel_size: 2
60 stride: 2
61 }
62 }
63 layers {
64 name: "conv2"
65 type: CONVOLUTION
66 bottom: "pool1"
67 top: "conv2"
68 blobs_lr: 1
69 blobs_lr: 2
70 convolution_param {
71 num_output: 50
72 kernel_size: 5
73 stride: 1
74 weight_filler {
75 type: "xavier"
76 }
77 bias_filler {
78 type: "constant"
79 }
80 }
81 }
82 layers {
83 name: "pool2"
84 type: POOLING
85 bottom: "conv2"
86 top: "pool2"
87 pooling_param {
88 pool: MAX
89 kernel_size: 2
90 stride: 2
91 }
92 }
93 layers {
94 name: "ip1"
95 type: INNER_PRODUCT
96 bottom: "pool2"
97 top: "ip1"
98 blobs_lr: 1
99 blobs_lr: 2
100 inner_product_param {
101 num_output: 500
102 weight_filler {
103 type: "xavier"
104 }
105 bias_filler {
106 type: "constant"
107 }
108 }
109 }
110 layers {
111 name: "relu1"
112 type: RELU
113 bottom: "ip1"
114 top: "ip1"
115 }
116 layers {
117 name: "ip2"
118 type: INNER_PRODUCT
119 bottom: "ip1"
120 top: "ip2"
121 blobs_lr: 1
122 blobs_lr: 2
123 inner_product_param {
124 num_output: 10
125 weight_filler {
126 type: "xavier"
127 }
128 bias_filler {
129 type: "constant"
130 }
131 }
132 }
133 layers {
134 name: "accuracy"
135 type: ACCURACY
136 bottom: "ip2"
137 bottom: "label"
138 top: "accuracy"
139 include: { phase: TEST }
140 }
141 layers {
142 name: "loss"
143 type: SOFTMAX_LOSS
144 bottom: "ip2"
145 bottom: "label"
146 top: "loss"
147 }
- 4、運行 ./example/mnist/train_lenet.sh
train_lenet.sh內容為
1 #!/usr/bin/env sh
2
3 ./build/tools/caffe train --solver=examples/mnist/lenet_solver.prototxt
它其實使用的是在lenet_solver.prototxt中定義的解決方案
lenet_solver.prototxt內容是
1 # The train/test net protocol buffer definition 2 net: "examples/mnist/lenet_train_test.prototxt" 3 # test_iter specifies how many forward passes the test should carry out. 4 # In the case of MNIST, we have test batch size 100 and 100 test iterations, 5 # covering the full 10,000 testing images. 6 test_iter: 100 7 # Carry out testing every 500 training iterations. 8 test_interval: 500 9 # The base learning rate, momentum and the weight decay of the network. 10 base_lr: 0.01 11 momentum: 0.9 12 weight_decay: 0.0005 13 # The learning rate policy 14 lr_policy: "inv" 15 gamma: 0.0001 16 power: 0.75 17 # Display every 100 iterations 18 display: 100 19 # The maximum number of iterations 20 max_iter: 10000 21 # snapshot intermediate results 22 snapshot: 5000 23 snapshot_prefix: "examples/mnist/lenet" 24 # solver mode: CPU or GPU 25 solver_mode: GPU
- 5.查看實驗結果：
經過一系列迭代iteration之后，在mnist上識別正確率能達到99.11%
