Image Recognition Tutorial in R MXNet Package

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

1. This is a detailed tutorial on image recognition in R using a deep convolutional neural network provided by the MXNet package. After a short post I wrote some times ago I received a lot of requests and emails for a much more detailed explanation, therefore I decided to write this tutorial. This post will show a reproducible example on how to get 97.5% accuracy score on a faces recognition task in R.

复制代码

本帖隐藏的内容

Image Recognition Tutorial in R MXNet Package.pdf (2.66 MB)

2017-4-26 03:01:45 上传

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏1 回帖

关键词：Recognition cognition Tutorial package image

1. # -*- coding: utf-8 -*-
   
2. 
   
3. # Imports
   
4. from sklearn.datasets import fetch_olivetti_faces
   
5. import numpy as np
   
6. 
   
7. # Download Olivetti faces dataset
   
8. olivetti = fetch_olivetti_faces()
   
9. x = olivetti.images
   
10. y = olivetti.target
    
11. 
    
12. # Print info on shapes and reshape where necessary
    
13. print("Original x shape:", x.shape)
    
14. X = x.reshape((400, 4096))
    
15. print("New x shape:", X.shape)
    
16. print("y shape", y.shape)
    
17. 
    
18. # Save the numpy arrays
    
19. np.savetxt("C://olivetti_X.csv", X, delimiter = ",")
    
20. np.savetxt("C://olivetti_y.csv", y, delimiter = ",", fmt = '%d')
    
21. 
    
22. print("\nDownloading and reshaping done!")

复制代码

1. # This script is used to resize images from 64x64 to 28x28 pixels
   
2. 
   
3. # Clear workspace
   
4. rm(list=ls())
   
5. 
   
6. # Load EBImage library
   
7. require(EBImage)
   
8. 
   
9. # Load data
   
10. X <- read.csv("olivetti_X.csv", header = F)
    
11. labels <- read.csv("olivetti_y.csv", header = F)
    
12. 
    
13. # Dataframe of resized images
    
14. rs_df <- data.frame()
    
15. 
    
16. # Main loop: for each image, resize and set it to greyscale
    
17. for(i in 1:nrow(X))
    
18. {
    
19.     # Try-catch
    
20.     result <- tryCatch({
    
21.     # Image (as 1d vector)
    
22.     img <- as.numeric(X[i,])
    
23.     # Reshape as a 64x64 image (EBImage object)
    
24.     img <- Image(img, dim=c(64, 64), colormode = "Grayscale")
    
25.     # Resize image to 28x28 pixels
    
26.     img_resized <- resize(img, w = 28, h = 28)
    
27.     # Get image matrix (there should be another function to do this faster and more neatly!)
    
28.     img_matrix <- img_resized@.Data
    
29.     # Coerce to a vector
    
30.     img_vector <- as.vector(t(img_matrix))
    
31.     # Add label
    
32.     label <- labels[i,]
    
33.     vec <- c(label, img_vector)
    
34.     # Stack in rs_df using rbind
    
35.     rs_df <- rbind(rs_df, vec)
    
36.     # Print status
    
37.     print(paste("Done",i,sep = " "))},
    
38.     # Error function (just prints the error). Btw you should get no errors!
    
39.     error = function(e){print(e)})
    
40. }
    
41. 
    
42. 
    
43. # Set names. The first columns are the labels, the other columns are the pixels.
    
44. names(rs_df) <- c("label", paste("pixel", c(1:784)))
    
45. 
    
46. # Train-test split
    
47. #-------------------------------------------------------------------------------
    
48. # Simple train-test split. No crossvalidation is done in this tutorial.
    
49. 
    
50. # Set seed for reproducibility purposes
    
51. set.seed(100)
    
52. 
    
53. # Shuffled df
    
54. shuffled <- rs_df[sample(1:400),]
    
55. 
    
56. # Train-test split
    
57. train_28 <- shuffled[1:360, ]
    
58. test_28 <- shuffled[361:400, ]
    
59. 
    
60. # Save train-test datasets
    
61. write.csv(train_28, "C://train_28.csv", row.names = FALSE)
    
62. write.csv(test_28, "C://test_28.csv", row.names = FALSE)
    
63. 
    
64. # Done!
    
65. print("Done!")

复制代码

1. # Clean workspace
   
2. rm(list=ls())
   
3. 
   
4. # Load MXNet
   
5. require(mxnet)
   
6. 
   
7. # Loading data and set up
   
8. #-------------------------------------------------------------------------------
   
9. 
   
10. # Load train and test datasets
    
11. train <- read.csv("train_28.csv")
    
12. test <- read.csv("test_28.csv")
    
13. 
    
14. # Set up train and test datasets
    
15. train <- data.matrix(train)
    
16. train_x <- t(train[, -1])
    
17. train_y <- train[, 1]
    
18. train_array <- train_x
    
19. dim(train_array) <- c(28, 28, 1, ncol(train_x))
    
20. 
    
21. test_x <- t(test[, -1])
    
22. test_y <- test[, 1]
    
23. test_array <- test_x
    
24. dim(test_array) <- c(28, 28, 1, ncol(test_x))
    
25. 
    
26. # Set up the symbolic model
    
27. #-------------------------------------------------------------------------------
    
28. 
    
29. data <- mx.symbol.Variable('data')
    
30. # 1st convolutional layer
    
31. conv_1 <- mx.symbol.Convolution(data = data, kernel = c(5, 5), num_filter = 20)
    
32. tanh_1 <- mx.symbol.Activation(data = conv_1, act_type = "tanh")
    
33. pool_1 <- mx.symbol.Pooling(data = tanh_1, pool_type = "max", kernel = c(2, 2), stride = c(2, 2))
    
34. # 2nd convolutional layer
    
35. conv_2 <- mx.symbol.Convolution(data = pool_1, kernel = c(5, 5), num_filter = 50)
    
36. tanh_2 <- mx.symbol.Activation(data = conv_2, act_type = "tanh")
    
37. pool_2 <- mx.symbol.Pooling(data=tanh_2, pool_type = "max", kernel = c(2, 2), stride = c(2, 2))
    
38. # 1st fully connected layer
    
39. flatten <- mx.symbol.Flatten(data = pool_2)
    
40. fc_1 <- mx.symbol.FullyConnected(data = flatten, num_hidden = 500)
    
41. tanh_3 <- mx.symbol.Activation(data = fc_1, act_type = "tanh")
    
42. # 2nd fully connected layer
    
43. fc_2 <- mx.symbol.FullyConnected(data = tanh_3, num_hidden = 40)
    
44. # Output. Softmax output since we'd like to get some probabilities.
    
45. NN_model <- mx.symbol.SoftmaxOutput(data = fc_2)
    
46. 
    
47. # Pre-training set up
    
48. #-------------------------------------------------------------------------------
    
49. 
    
50. # Set seed for reproducibility
    
51. mx.set.seed(100)
    
52. 
    
53. # Device used. CPU in my case.
    
54. devices <- mx.cpu()
    
55. 
    
56. # Training
    
57. #-------------------------------------------------------------------------------
    
58. 
    
59. # Train the model
    
60. model <- mx.model.FeedForward.create(NN_model,
    
61.                                      X = train_array,
    
62.                                      y = train_y,
    
63.                                      ctx = devices,
    
64.                                      num.round = 480,
    
65.                                      array.batch.size = 40,
    
66.                                      learning.rate = 0.01,
    
67.                                      momentum = 0.9,
    
68.                                      eval.metric = mx.metric.accuracy,
    
69.                                      epoch.end.callback = mx.callback.log.train.metric(100))
    
70. 
    
71. # Testing
    
72. #-------------------------------------------------------------------------------
    
73. 
    
74. # Predict labels
    
75. predicted <- predict(model, test_array)
    
76. # Assign labels
    
77. predicted_labels <- max.col(t(predicted)) - 1
    
78. # Get accuracy
    
79. sum(diag(table(test[, 1], predicted_labels)))/40

复制代码

https://firsttimeprogrammer.blogspot.ca/2016/08/image-recognition-tutorial-in-r-using.html

1. # Clean workspace
   
2. rm(list=ls())
   
3. 
   
4. # Load MXNet
   
5. require(mxnet)
   
6. 
   
7. # Loading data and set up
   
8. #-------------------------------------------------------------------------------
   
9. 
   
10. # Load train and test datasets
    
11. train <- read.csv("train_28.csv")
    
12. test <- read.csv("test_28.csv")
    
13. 
    
14. # Set up train and test datasets
    
15. train <- data.matrix(train)
    
16. train_x <- t(train[, -1])
    
17. train_y <- train[, 1]
    
18. train_array <- train_x
    
19. dim(train_array) <- c(28, 28, 1, ncol(train_x))
    
20. 
    
21. test_x <- t(test[, -1])
    
22. test_y <- test[, 1]
    
23. test_array <- test_x
    
24. dim(test_array) <- c(28, 28, 1, ncol(test_x))
    
25. 
    
26. # Set up the symbolic model
    
27. #-------------------------------------------------------------------------------
    
28. 
    
29. data <- mx.symbol.Variable('data')
    
30. # 1st convolutional layer
    
31. conv_1 <- mx.symbol.Convolution(data = data, kernel = c(5, 5), num_filter = 20)
    
32. tanh_1 <- mx.symbol.Activation(data = conv_1, act_type = "tanh")
    
33. pool_1 <- mx.symbol.Pooling(data = tanh_1, pool_type = "max", kernel = c(2, 2), stride = c(2, 2))
    
34. # 2nd convolutional layer
    
35. conv_2 <- mx.symbol.Convolution(data = pool_1, kernel = c(5, 5), num_filter = 50)
    
36. tanh_2 <- mx.symbol.Activation(data = conv_2, act_type = "tanh")
    
37. pool_2 <- mx.symbol.Pooling(data=tanh_2, pool_type = "max", kernel = c(2, 2), stride = c(2, 2))
    
38. # 1st fully connected layer
    
39. flatten <- mx.symbol.Flatten(data = pool_2)
    
40. fc_1 <- mx.symbol.FullyConnected(data = flatten, num_hidden = 500)
    
41. tanh_3 <- mx.symbol.Activation(data = fc_1, act_type = "tanh")
    
42. # 2nd fully connected layer
    
43. fc_2 <- mx.symbol.FullyConnected(data = tanh_3, num_hidden = 40)
    
44. # Output. Softmax output since we'd like to get some probabilities.
    
45. NN_model <- mx.symbol.SoftmaxOutput(data = fc_2)
    
46. 
    
47. # Pre-training set up
    
48. #-------------------------------------------------------------------------------
    
49. 
    
50. # Set seed for reproducibility
    
51. mx.set.seed(100)
    
52. 
    
53. # Device used. CPU in my case.
    
54. devices <- mx.cpu()
    
55. 
    
56. # Training
    
57. #-------------------------------------------------------------------------------
    
58. 
    
59. # Train the model
    
60. model <- mx.model.FeedForward.create(NN_model,
    
61.                                      X = train_array,
    
62.                                      y = train_y,
    
63.                                      ctx = devices,
    
64.                                      num.round = 480,
    
65.                                      array.batch.size = 40,
    
66.                                      learning.rate = 0.01,
    
67.                                      momentum = 0.9,
    
68.                                      eval.metric = mx.metric.accuracy,
    
69.                                      epoch.end.callback = mx.callback.log.train.metric(100))
    
70. 
    
71. # Testing
    
72. #-------------------------------------------------------------------------------
    
73. 
    
74. # Predict labels
    
75. predicted <- predict(model, test_array)
    
76. # Assign labels
    
77. predicted_labels <- max.col(t(predicted)) - 1
    
78. # Get accuracy
    
79. sum(diag(table(test[, 1], predicted_labels)))/40
    
80. 
    
81. ################################################################################
    
82. #                           OUTPUT
    
83. ################################################################################
    
84. #
    
85. # 0.975
    
86. #

复制代码

thanks for sharing

Image Recognition Tutorial in R MXNet Package