classification_cnn_keras.py 6.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244
  1. '''This script goes along the blog post
  2. "Building powerful image classification models using very little data"
  3. from blog.keras.io.
  4. ```
  5. data/
  6. train/
  7. final/
  8. final001.png
  9. final002.png
  10. ...
  11. noisy/
  12. noisy001.png
  13. noisy002.png
  14. ...
  15. validation/
  16. final/
  17. final001.png
  18. final002.png
  19. ...
  20. noisy/
  21. noisy001.png
  22. noisy002.png
  23. ...
  24. ```
  25. '''
  26. import sys, os, getopt
  27. import json
  28. from keras.preprocessing.image import ImageDataGenerator
  29. from keras.models import Sequential
  30. from keras.layers import Conv2D, MaxPooling2D, AveragePooling2D
  31. from keras.layers import Activation, Dropout, Flatten, Dense, BatchNormalization
  32. from keras import backend as K
  33. from keras.utils import plot_model
  34. from ipfml import tf_model_helper
  35. # local functions import (metrics preprocessing)
  36. import preprocessing_functions
  37. ##########################################
  38. # Global parameters (with default value) #
  39. ##########################################
  40. img_width, img_height = 100, 100
  41. train_data_dir = 'data/train'
  42. validation_data_dir = 'data/validation'
  43. nb_train_samples = 7200
  44. nb_validation_samples = 3600
  45. epochs = 50
  46. batch_size = 16
  47. input_shape = (3, img_width, img_height)
  48. ###########################################
  49. '''
  50. Method which returns model to train
  51. @return : DirectoryIterator
  52. '''
  53. def generate_model():
  54. model = Sequential()
  55. model.add(Conv2D(60, (2, 2), input_shape=input_shape))
  56. model.add(Activation('relu'))
  57. model.add(MaxPooling2D(pool_size=(2, 2)))
  58. model.add(Conv2D(40, (2, 2)))
  59. model.add(Activation('relu'))
  60. model.add(MaxPooling2D(pool_size=(2, 2)))
  61. model.add(Conv2D(20, (2, 2)))
  62. model.add(Activation('relu'))
  63. model.add(MaxPooling2D(pool_size=(2, 2)))
  64. model.add(Flatten())
  65. model.add(Dense(140))
  66. model.add(Activation('relu'))
  67. model.add(BatchNormalization())
  68. model.add(Dropout(0.3))
  69. model.add(Dense(120))
  70. model.add(Activation('relu'))
  71. model.add(BatchNormalization())
  72. model.add(Dropout(0.3))
  73. model.add(Dense(80))
  74. model.add(Activation('relu'))
  75. model.add(BatchNormalization())
  76. model.add(Dropout(0.2))
  77. model.add(Dense(40))
  78. model.add(Activation('relu'))
  79. model.add(BatchNormalization())
  80. model.add(Dropout(0.2))
  81. model.add(Dense(20))
  82. model.add(Activation('relu'))
  83. model.add(BatchNormalization())
  84. model.add(Dropout(0.2))
  85. model.add(Dense(1))
  86. model.add(Activation('sigmoid'))
  87. model.compile(loss='binary_crossentropy',
  88. optimizer='rmsprop',
  89. metrics=['accuracy'])
  90. return model
  91. '''
  92. Method which loads train data
  93. @return : DirectoryIterator
  94. '''
  95. def load_train_data():
  96. # this is the augmentation configuration we will use for training
  97. train_datagen = ImageDataGenerator(
  98. rescale=1. / 255,
  99. shear_range=0.2,
  100. zoom_range=0.2,
  101. horizontal_flip=True)
  102. train_generator = train_datagen.flow_from_directory(
  103. train_data_dir,
  104. target_size=(img_width, img_height),
  105. batch_size=batch_size,
  106. class_mode='binary')
  107. return train_generator
  108. '''
  109. Method which loads validation data
  110. @return : DirectoryIterator
  111. '''
  112. def load_validation_data():
  113. # this is the augmentation configuration we will use for testing:
  114. # only rescaling
  115. test_datagen = ImageDataGenerator(rescale=1. / 255)
  116. validation_generator = test_datagen.flow_from_directory(
  117. validation_data_dir,
  118. target_size=(img_width, img_height),
  119. batch_size=batch_size,
  120. class_mode='binary')
  121. return validation_generator
  122. def main():
  123. # update global variable and not local
  124. global batch_size
  125. global epochs
  126. global img_width
  127. global img_height
  128. global input_shape
  129. global train_data_dir
  130. global validation_data_dir
  131. global nb_train_samples
  132. global nb_validation_samples
  133. if len(sys.argv) <= 1:
  134. print('Run with default parameters...')
  135. print('classification_cnn_keras_svd.py --directory xxxx --output xxxxx --batch_size xx --epochs xx --img xx')
  136. sys.exit(2)
  137. try:
  138. opts, args = getopt.getopt(sys.argv[1:], "ho:d:b:e:i", ["help", "output=", "directory=", "batch_size=", "epochs=", "img="])
  139. except getopt.GetoptError:
  140. # print help information and exit:
  141. print('classification_cnn_keras_svd.py --directory xxxx --output xxxxx --batch_size xx --epochs xx --img xx')
  142. sys.exit(2)
  143. for o, a in opts:
  144. if o == "-h":
  145. print('classification_cnn_keras_svd.py --directory xxxx --output xxxxx --batch_size xx --epochs xx --img xx')
  146. sys.exit()
  147. elif o in ("-o", "--output"):
  148. filename = a
  149. elif o in ("-b", "--batch_size"):
  150. batch_size = int(a)
  151. print(batch_size)
  152. elif o in ("-e", "--epochs"):
  153. epochs = int(a)
  154. elif o in ("-d", "--directory"):
  155. directory = a
  156. elif o in ("-i", "--img"):
  157. img_height = int(a)
  158. img_width = int(a)
  159. else:
  160. assert False, "unhandled option"
  161. # 3 because we have 3 color canals
  162. if K.image_data_format() == 'channels_first':
  163. input_shape = (3, img_width, img_height)
  164. else:
  165. input_shape = (img_width, img_height, 3)
  166. # configuration
  167. with open('config.json') as json_data:
  168. d = json.load(json_data)
  169. train_data_dir = d['train_data_dir']
  170. validation_data_dir = d['train_validation_dir']
  171. try:
  172. nb_train_samples = d[str(img_width)]['nb_train_samples']
  173. nb_validation_samples = d[str(img_width)]['nb_validation_samples']
  174. except:
  175. print("--img parameter missing of invalid (--image_width xx --img_height xx)")
  176. sys.exit(2)
  177. # load of model
  178. model = generate_model()
  179. model.summary()
  180. if 'directory' in locals():
  181. print('Your model information will be saved into %s...' % directory)
  182. history = model.fit_generator(
  183. load_train_data(),
  184. steps_per_epoch=nb_train_samples // batch_size,
  185. epochs=epochs,
  186. validation_data=load_validation_data(),
  187. validation_steps=nb_validation_samples // batch_size)
  188. # if user needs output files
  189. if(filename):
  190. # update filename by folder
  191. if(directory):
  192. # create folder if necessary
  193. if not os.path.exists(directory):
  194. os.makedirs(directory)
  195. filename = directory + "/" + filename
  196. # save plot file history
  197. tf_model_helper.save(history, filename)
  198. plot_model(model, to_file=str(('%s.png' % filename)))
  199. model.save_weights(str('%s.h5' % filename))
  200. if __name__ == "__main__":
  201. main()