Caffe实现概述 - 润新知

Caffe实现概述
Caffe实现概述

目录

一、caffe配置文件介绍

二、标准层的定义

三、网络微调技巧

四、Linux脚本使用及LMDB文件生成

五、带你设计一个Caffe网络，用于分类任务

一、caffe配置文件介绍

二、标准层的定义

三、网络微调技巧

其中，multistep最为常用

四、Linux脚本使用及LMDB文件生成

五、带你设计一个Caffe网络，用于分类任务

下面：

使用pycaffe生成solver配置

使用pycaffe生成caffe测试网络和训练网络

数据集下载：

# demoCaffe

数据集下载，cifar mnist:
百度云盘:

链接: https://pan.baidu.com/s/1bHFQUz7Q6BMBZv25AhsXKQ 密码: dva9
链接: https://pan.baidu.com/s/1rPRjf2hanlYYjBQQDmIjNQ 密码: 5nhv

1. lmdb数据制作：

手动实现： https://blog.csdn.net/yx2017/article/details/72953537

https://www.jianshu.com/p/9d7ed35960cb

代码实现：https://www.cnblogs.com/leemo-o/p/4990021.html

https://www.jianshu.com/p/ef84715e0fdc

以下仅供对比阅读：

demo_lmdb.py：生成lmdb格式数据
1. import lmdb
2. import numpy as np
3. import cv2
4. import caffe
5. from caffe.proto import caffe_pb2
7. def write():
8. # basic setting
10. 10. lmdb_file = 'lmdb_data'
11. 11. batch_size = 256
12. 12.
13. 13.
14. 14. lmdb_env = lmdb.open(lmdb_file, map_size = int(1e12))
15. 15.
16. 16. lmdb_txn = lmdb_env.begin(write = True)
17. 17.
18. 18. for x in range(batch_size):
19. 19. data = np.ones((3, 64, 64), np.uint8)
20. 20. label = x
21. 21.
22. 22. datum = caffe.io.array_to_datum(data,label)
23. 23. keystr = "{:0>8d}".format(x)
24. 24.
25. 25. lmdb_txn.put(keystr, datum.SerializeToString())
26. 26.
27. 27. lmdb_txn.commit()
28. 28.
29. def read():
1. 30. lmdb_env = lmdb.open('lmdb_data')
2. 31. lmdb_txt = lmdb_env.begin()
3. 32.
4. 33. datum = caffe_pb2.Datum()
5. 34.
6. 35. for key, value in lmdb_txt.cursor():
7. 36.
8. 37. datum.ParseFromString(value)
9. 38.
10. 39. label = datum.label
11. 40.
12. 41. data = caffe.io.datum_to_array(datum)
13. 42.
14. 43. print(label)
15. 44. print(data)
16. 45.
17. 46.
47. if __name__ == '__main__':
1. 48. write()
2. 49. read()
demo_create_solver.py: 生成solver配置文件
1. from caffe.proto import caffe_pb2
3. s = caffe_pb2.SolverParameter()
5. s.train_net = "train.prototxt"
6. s.test_net.append("test.prototxt")
8. s.test_interval = 100
9. s.test_iter.append(10)
10. 10.
11. s.max_iter = 1000
1. 12.
13. s.base_lr = 0.1
1. 14.
15. s.weight_decay = 5e-4
1. 16.
17. s.lr_policy = "step"
1. 18.
19. s.display = 10
1. 20.
21. s.snapshot = 10
1. 22.
23. s.snapshot_prefix = "model"
1. 24.
25. s.type = "SGD"
1. 26.
27. s.solver_mode = caffe_pb2.SolverParameter.GPU
1. 28.
29. with open("net/s.prototxt", "w") as f:
1. 30. f.write(str(s))
2. 31.
3. 32.
4. 33.
5. 34.
结果如下：
1. train_net: "/home/kuan/PycharmProjects/demo_cnn_net/net/train.prototxt"
2. test_net: "/home/kuan/PycharmProjects/demo_cnn_net/net/test.prototxt"
3. test_iter: 1000
4. test_interval: 100
5. base_lr: 0.10000000149
6. display: 100
7. max_iter: 100000
8. lr_policy: "step"
9. weight_decay: 0.000500000023749
10. snapshot: 100

11. snapshot_prefix: "/home/kuan/PycharmProjects/demo_cnn_net/cnn_model/mnist/lenet/"

12. solver_mode: GPU

13. type: "SGD"
demo_creat_net.py: 创建网络
1. import caffe
3. def create_net():
4. net = caffe.NetSpec()
6. net.data, net.label = caffe.layers.Data(source="data.lmdb",
7. backend=caffe.params.Data.LMDB,
8. batch_size=32,
9. ntop=2, #数据层数据个数，分别为data,label
10. 10. transform_param=dict(crop_size=40, mirror=True)
11. 11. )
12. 12.
13. 13. net.conv1 = caffe.layers.Convolution(net.data, num_output=20, kernel_size=5,
14. 14. weight_filler={"type": "xavier"},
15. 15. bias_filler={"type":"xavier"}) #卷积核参数
16. 16.
17. 17. net.relu1 = caffe.layers.ReLU(net.conv1, in_place=True)
18. 18.
19. 19. net.pool1 = caffe.layers.Pooling(net.relu1, pool=caffe.params.Pooling.MAX,
20. 20. kernel_size=3, stride=2)
21. 21.
22. 22. net.conv2 = caffe.layers.Convolution(net.pool1, num_output=32, kernel_size=3,
23. 23. pad=1,
24. 24. weight_filler={"type": "xavier"},
25. 25. bias_filler={"type": "xavier"})
26. 26.
27. 27. net.relu2 = caffe.layers.ReLU(net.conv2, in_place=True)
28. 28.
29. 29. net.pool2 = caffe.layers.Pooling(net.relu2, pool=caffe.params.Pooling.MAX,
30. 30. kernel_size=3, stride=2)
31. 31. #下面为全连接层
32. 32. net.fc3 = caffe.layers.InnerProduct(net.pool2, num_output=1024, weight_filler=dict(type='xavier'))
33. 33.
34. 34. net.relu3 = caffe.layers.ReLU(net.fc3, in_place=True)
35. 35.
36. 36. ##drop
37. 37. net.drop = caffe.layers.Dropout(net.relu3, dropout_param=dict(dropout_ratio=0.5))
38. 38.
39. 39. net.fc4 = caffe.layers.InnerProduct(net.drop, num_output=10, weight_filler=dict(type='xavier'))
40. 40.
41. 41. net.loss = caffe.layers.SoftmaxWithLoss(net.fc4, net.label)
42. 42.
43. 43. with open("net/tt.prototxt", 'w') as f:
44. 44. f.write(str(net.to_proto()))
45. 45.
46. 46.
47. if __name__ == '__main__':
1. 48. create_net()
生成结果如下：
1. layer {
2. name: "data"
3. type: "Data"
4. top: "data"
5. top: "label"
6. transform_param {
7. mirror: true
8. crop_size: 40
9. }
10. 10. data_param {
11. 11. source: "/home/kuan/PycharmProjects/demo_cnn_net/lmdb_data"
12. 12. batch_size: 32
13. 13. backend: LMDB
14. 14. }
15. }

16. layer {
1. 17. name: "conv1"
2. 18. type: "Convolution"
3. 19. bottom: "data"
4. 20. top: "conv1"
5. 21. convolution_param {
6. 22. num_output: 20
7. 23. kernel_size: 5
8. 24. weight_filler {
9. 25. type: "xavier"
10. 26. }
11. 27. bias_filler {
12. 28. type: "xavier"
13. 29. }
14. 30. }
31. }

32. layer {
1. 33. name: "relu1"
2. 34. type: "ReLU"
3. 35. bottom: "conv1"
4. 36. top: "conv1"
37. }

38. layer {
1. 39. name: "pool1"
2. 40. type: "Pooling"
3. 41. bottom: "conv1"
4. 42. top: "pool1"
5. 43. pooling_param {
6. 44. pool: MAX
7. 45. kernel_size: 3
8. 46. stride: 2
9. 47. }
48. }

49. layer {
1. 50. name: "conv2"
2. 51. type: "Convolution"
3. 52. bottom: "pool1"
4. 53. top: "conv2"
5. 54. convolution_param {
6. 55. num_output: 32
7. 56. pad: 1
8. 57. kernel_size: 3
9. 58. weight_filler {
10. 59. type: "xavier"
11. 60. }
12. 61. bias_filler {
13. 62. type: "xavier"
14. 63. }
15. 64. }
65. }

66. layer {
1. 67. name: "relu2"
2. 68. type: "ReLU"
3. 69. bottom: "conv2"
4. 70. top: "conv2"
71. }

72. layer {
1. 73. name: "pool2"
2. 74. type: "Pooling"
3. 75. bottom: "conv2"
4. 76. top: "pool2"
5. 77. pooling_param {
6. 78. pool: MAX
7. 79. kernel_size: 3
8. 80. stride: 2
9. 81. }
82. }

83. layer {
1. 84. name: "fc3"
2. 85. type: "InnerProduct"
3. 86. bottom: "pool2"
4. 87. top: "fc3"
5. 88. inner_product_param {
6. 89. num_output: 1024
7. 90. weight_filler {
8. 91. type: "xavier"
9. 92. }
10. 93. }
94. }

95. layer {
1. 96. name: "relu3"
2. 97. type: "ReLU"
3. 98. bottom: "fc3"
4. 99. top: "fc3"
5. }
6. layer {
7. name: "drop"
8. type: "Dropout"
9. bottom: "fc3"
10. top: "drop"
11. dropout_param {
12. dropout_ratio: 0.5
13. }
14. }
15. layer {
16. name: "fc4"
17. type: "InnerProduct"
18. bottom: "drop"
19. top: "fc4"
20. inner_product_param {
21. num_output: 10
22. weight_filler {
23. type: "xavier"
24. }
25. }
26. }
27. layer {
28. name: "loss"
29. type: "SoftmaxWithLoss"
30. bottom: "fc4"
31. bottom: "label"
32. top: "loss"
33. }
demo_train.py训练网络：
1. import sys
2. sys.path.append('/home/kuan/AM-softmax_caffe/python')
3. import caffe
5. solver = caffe.SGDSolver("/home/kuan/PycharmProjects/demo_cnn_net/cnn_net/alexnet/solver.prototxt")
7. solver.solve()
demo_test.py:测试网络
1. import sys
2. sys.path.append('/home/kuan/AM-softmax_caffe/python')
3. import caffe
4. import numpy as np
6. ##caffemodel deploy.prototxt
8. deploy = "/home/kuan/PycharmProjects/demo_cnn_net/cnn_net/alexnet/deploy.prototxt"
10. model = "/home/kuan/PycharmProjects/demo_cnn_net/cnn_model/cifar/alexnet/alexnet_iter_110.caffemodel"
1. 11.
12. net = caffe.Net(deploy, model, caffe.TEST)
1. 13.
2. 14.
15. net.blobs["data"].data[...] = np.ones((3,32,32),np.uint8)
1. 16.
17. net.forward()
1. 18.
19. prob = net.blobs["prob"].data[0]
1. 20.
21. print(prob)
1. 22.
人工智能芯片与自动驾驶
相关阅读:
tomcat启动项目，起不起来
 jQuery ajax 获取信息展示在“下拉列表”中
 JQuery,ajax,jsonp 跨域访问
 httpwebrequest详解
 利用HttpWebRequest和HttpWebResponse获取Cookie并实现模拟登录
 C# 中使用System.Net.Http.HttpClient 模拟登录博客园 (GET/POST)
JAVA-用HttpClient来模拟浏览器GET,POST
java 实现模拟浏览器访问网站
 ora-01017 和oracle database link
JDeveloper 开发环境配置
原文地址：https://www.cnblogs.com/wujianming-110117/p/14391739.html