API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Aug 26th, 2019
130
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 17.63 KB | None | 0 0
raw download clone embed print report
  1. from __future__ import print_function
  2. import imageio
  3. from PIL import Image
  4. import numpy as np
  5. import keras
  6.  
  7. from keras.layers import Input, Dense, Conv2D, MaxPooling2D, AveragePooling2D, ZeroPadding2D, Dropout, Flatten, Concatenate, Reshape, Activation
  8. from keras.models import Model
  9. from keras.regularizers import l2
  10. from keras.optimizers import SGD
  11. from pool_helper import PoolHelper
  12. from lrn import LRN
  13.  
  14. if keras.backend.backend() == 'tensorflow':
  15. from keras import backend as K
  16. import tensorflow as tf
  17. from keras.utils.conv_utils import convert_kernel
  18.  
  19.  
  20. def create_googlenet(weights_path=None):
  21. # creates GoogLeNet a.k.a. Inception v1 (Szegedy, 2015)
  22. input = Input(shape=(3, 224, 224))
  23.  
  24. input_pad = ZeroPadding2D(padding=(3, 3))(input)
  25. conv1_7x7_s2 = Conv2D(64, (7,7), strides=(2,2), padding='valid', activation='relu', name='conv1/7x7_s2', kernel_regularizer=l2(0.0002))(input_pad)
  26. conv1_zero_pad = ZeroPadding2D(padding=(1, 1))(conv1_7x7_s2)
  27. pool1_helper = PoolHelper()(conv1_zero_pad)
  28. pool1_3x3_s2 = MaxPooling2D(pool_size=(3,3), strides=(2,2), padding='valid', name='pool1/3x3_s2')(pool1_helper)
  29. pool1_norm1 = LRN(name='pool1/norm1')(pool1_3x3_s2)
  30.  
  31. conv2_3x3_reduce = Conv2D(64, (1,1), padding='same', activation='relu', name='conv2/3x3_reduce', kernel_regularizer=l2(0.0002))(pool1_norm1)
  32. conv2_3x3 = Conv2D(192, (3,3), padding='same', activation='relu', name='conv2/3x3', kernel_regularizer=l2(0.0002))(conv2_3x3_reduce)
  33. conv2_norm2 = LRN(name='conv2/norm2')(conv2_3x3)
  34. conv2_zero_pad = ZeroPadding2D(padding=(1, 1))(conv2_norm2)
  35. pool2_helper = PoolHelper()(conv2_zero_pad)
  36. pool2_3x3_s2 = MaxPooling2D(pool_size=(3,3), strides=(2,2), padding='valid', name='pool2/3x3_s2')(pool2_helper)
  37.  
  38. inception_3a_1x1 = Conv2D(64, (1,1), padding='same', activation='relu', name='inception_3a/1x1', kernel_regularizer=l2(0.0002))(pool2_3x3_s2)
  39. inception_3a_3x3_reduce = Conv2D(96, (1,1), padding='same', activation='relu', name='inception_3a/3x3_reduce', kernel_regularizer=l2(0.0002))(pool2_3x3_s2)
  40. inception_3a_3x3_pad = ZeroPadding2D(padding=(1, 1))(inception_3a_3x3_reduce)
  41. inception_3a_3x3 = Conv2D(128, (3,3), padding='valid', activation='relu', name='inception_3a/3x3', kernel_regularizer=l2(0.0002))(inception_3a_3x3_pad)
  42. inception_3a_5x5_reduce = Conv2D(16, (1,1), padding='same', activation='relu', name='inception_3a/5x5_reduce', kernel_regularizer=l2(0.0002))(pool2_3x3_s2)
  43. inception_3a_5x5_pad = ZeroPadding2D(padding=(2, 2))(inception_3a_5x5_reduce)
  44. inception_3a_5x5 = Conv2D(32, (5,5), padding='valid', activation='relu', name='inception_3a/5x5', kernel_regularizer=l2(0.0002))(inception_3a_5x5_pad)
  45. inception_3a_pool = MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same', name='inception_3a/pool')(pool2_3x3_s2)
  46. inception_3a_pool_proj = Conv2D(32, (1,1), padding='same', activation='relu', name='inception_3a/pool_proj', kernel_regularizer=l2(0.0002))(inception_3a_pool)
  47. inception_3a_output = Concatenate(axis=1, name='inception_3a/output')([inception_3a_1x1,inception_3a_3x3,inception_3a_5x5,inception_3a_pool_proj])
  48.  
  49. inception_3b_1x1 = Conv2D(128, (1,1), padding='same', activation='relu', name='inception_3b/1x1', kernel_regularizer=l2(0.0002))(inception_3a_output)
  50. inception_3b_3x3_reduce = Conv2D(128, (1,1), padding='same', activation='relu', name='inception_3b/3x3_reduce', kernel_regularizer=l2(0.0002))(inception_3a_output)
  51. inception_3b_3x3_pad = ZeroPadding2D(padding=(1, 1))(inception_3b_3x3_reduce)
  52. inception_3b_3x3 = Conv2D(192, (3,3), padding='valid', activation='relu', name='inception_3b/3x3', kernel_regularizer=l2(0.0002))(inception_3b_3x3_pad)
  53. inception_3b_5x5_reduce = Conv2D(32, (1,1), padding='same', activation='relu', name='inception_3b/5x5_reduce', kernel_regularizer=l2(0.0002))(inception_3a_output)
  54. inception_3b_5x5_pad = ZeroPadding2D(padding=(2, 2))(inception_3b_5x5_reduce)
  55. inception_3b_5x5 = Conv2D(96, (5,5), padding='valid', activation='relu', name='inception_3b/5x5', kernel_regularizer=l2(0.0002))(inception_3b_5x5_pad)
  56. inception_3b_pool = MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same', name='inception_3b/pool')(inception_3a_output)
  57. inception_3b_pool_proj = Conv2D(64, (1,1), padding='same', activation='relu', name='inception_3b/pool_proj', kernel_regularizer=l2(0.0002))(inception_3b_pool)
  58. inception_3b_output = Concatenate(axis=1, name='inception_3b/output')([inception_3b_1x1,inception_3b_3x3,inception_3b_5x5,inception_3b_pool_proj])
  59.  
  60. inception_3b_output_zero_pad = ZeroPadding2D(padding=(1, 1))(inception_3b_output)
  61. pool3_helper = PoolHelper()(inception_3b_output_zero_pad)
  62. pool3_3x3_s2 = MaxPooling2D(pool_size=(3,3), strides=(2,2), padding='valid', name='pool3/3x3_s2')(pool3_helper)
  63.  
  64. inception_4a_1x1 = Conv2D(192, (1,1), padding='same', activation='relu', name='inception_4a/1x1', kernel_regularizer=l2(0.0002))(pool3_3x3_s2)
  65. inception_4a_3x3_reduce = Conv2D(96, (1,1), padding='same', activation='relu', name='inception_4a/3x3_reduce', kernel_regularizer=l2(0.0002))(pool3_3x3_s2)
  66. inception_4a_3x3_pad = ZeroPadding2D(padding=(1, 1))(inception_4a_3x3_reduce)
  67. inception_4a_3x3 = Conv2D(208, (3,3), padding='valid', activation='relu', name='inception_4a/3x3' ,kernel_regularizer=l2(0.0002))(inception_4a_3x3_pad)
  68. inception_4a_5x5_reduce = Conv2D(16, (1,1), padding='same', activation='relu', name='inception_4a/5x5_reduce', kernel_regularizer=l2(0.0002))(pool3_3x3_s2)
  69. inception_4a_5x5_pad = ZeroPadding2D(padding=(2, 2))(inception_4a_5x5_reduce)
  70. inception_4a_5x5 = Conv2D(48, (5,5), padding='valid', activation='relu', name='inception_4a/5x5', kernel_regularizer=l2(0.0002))(inception_4a_5x5_pad)
  71. inception_4a_pool = MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same', name='inception_4a/pool')(pool3_3x3_s2)
  72. inception_4a_pool_proj = Conv2D(64, (1,1), padding='same', activation='relu', name='inception_4a/pool_proj', kernel_regularizer=l2(0.0002))(inception_4a_pool)
  73. inception_4a_output = Concatenate(axis=1, name='inception_4a/output')([inception_4a_1x1,inception_4a_3x3,inception_4a_5x5,inception_4a_pool_proj])
  74.  
  75. loss1_ave_pool = AveragePooling2D(pool_size=(5,5), strides=(3,3), name='loss1/ave_pool')(inception_4a_output)
  76. loss1_conv = Conv2D(128, (1,1), padding='same', activation='relu', name='loss1/conv', kernel_regularizer=l2(0.0002))(loss1_ave_pool)
  77. loss1_flat = Flatten()(loss1_conv)
  78. loss1_fc = Dense(1024, activation='relu', name='loss1/fc', kernel_regularizer=l2(0.0002))(loss1_flat)
  79. loss1_drop_fc = Dropout(rate=0.7)(loss1_fc)
  80. loss1_classifier = Dense(1000, name='loss1/classifier', kernel_regularizer=l2(0.0002))(loss1_drop_fc)
  81. loss1_classifier_act = Activation('softmax')(loss1_classifier)
  82.  
  83. inception_4b_1x1 = Conv2D(160, (1,1), padding='same', activation='relu', name='inception_4b/1x1', kernel_regularizer=l2(0.0002))(inception_4a_output)
  84. inception_4b_3x3_reduce = Conv2D(112, (1,1), padding='same', activation='relu', name='inception_4b/3x3_reduce', kernel_regularizer=l2(0.0002))(inception_4a_output)
  85. inception_4b_3x3_pad = ZeroPadding2D(padding=(1, 1))(inception_4b_3x3_reduce)
  86. inception_4b_3x3 = Conv2D(224, (3,3), padding='valid', activation='relu', name='inception_4b/3x3', kernel_regularizer=l2(0.0002))(inception_4b_3x3_pad)
  87. inception_4b_5x5_reduce = Conv2D(24, (1,1), padding='same', activation='relu', name='inception_4b/5x5_reduce', kernel_regularizer=l2(0.0002))(inception_4a_output)
  88. inception_4b_5x5_pad = ZeroPadding2D(padding=(2, 2))(inception_4b_5x5_reduce)
  89. inception_4b_5x5 = Conv2D(64, (5,5), padding='valid', activation='relu', name='inception_4b/5x5', kernel_regularizer=l2(0.0002))(inception_4b_5x5_pad)
  90. inception_4b_pool = MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same', name='inception_4b/pool')(inception_4a_output)
  91. inception_4b_pool_proj = Conv2D(64, (1,1), padding='same', activation='relu', name='inception_4b/pool_proj', kernel_regularizer=l2(0.0002))(inception_4b_pool)
  92. inception_4b_output = Concatenate(axis=1, name='inception_4b/output')([inception_4b_1x1,inception_4b_3x3,inception_4b_5x5,inception_4b_pool_proj])
  93.  
  94. inception_4c_1x1 = Conv2D(128, (1,1), padding='same', activation='relu', name='inception_4c/1x1', kernel_regularizer=l2(0.0002))(inception_4b_output)
  95. inception_4c_3x3_reduce = Conv2D(128, (1,1), padding='same', activation='relu', name='inception_4c/3x3_reduce', kernel_regularizer=l2(0.0002))(inception_4b_output)
  96. inception_4c_3x3_pad = ZeroPadding2D(padding=(1, 1))(inception_4c_3x3_reduce)
  97. inception_4c_3x3 = Conv2D(256, (3,3), padding='valid', activation='relu', name='inception_4c/3x3', kernel_regularizer=l2(0.0002))(inception_4c_3x3_pad)
  98. inception_4c_5x5_reduce = Conv2D(24, (1,1), padding='same', activation='relu', name='inception_4c/5x5_reduce', kernel_regularizer=l2(0.0002))(inception_4b_output)
  99. inception_4c_5x5_pad = ZeroPadding2D(padding=(2, 2))(inception_4c_5x5_reduce)
  100. inception_4c_5x5 = Conv2D(64, (5,5), padding='valid', activation='relu', name='inception_4c/5x5', kernel_regularizer=l2(0.0002))(inception_4c_5x5_pad)
  101. inception_4c_pool = MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same', name='inception_4c/pool')(inception_4b_output)
  102. inception_4c_pool_proj = Conv2D(64, (1,1), padding='same', activation='relu', name='inception_4c/pool_proj', kernel_regularizer=l2(0.0002))(inception_4c_pool)
  103. inception_4c_output = Concatenate(axis=1, name='inception_4c/output')([inception_4c_1x1,inception_4c_3x3,inception_4c_5x5,inception_4c_pool_proj])
  104.  
  105. inception_4d_1x1 = Conv2D(112, (1,1), padding='same', activation='relu', name='inception_4d/1x1', kernel_regularizer=l2(0.0002))(inception_4c_output)
  106. inception_4d_3x3_reduce = Conv2D(144, (1,1), padding='same', activation='relu', name='inception_4d/3x3_reduce', kernel_regularizer=l2(0.0002))(inception_4c_output)
  107. inception_4d_3x3_pad = ZeroPadding2D(padding=(1, 1))(inception_4d_3x3_reduce)
  108. inception_4d_3x3 = Conv2D(288, (3,3), padding='valid', activation='relu', name='inception_4d/3x3', kernel_regularizer=l2(0.0002))(inception_4d_3x3_pad)
  109. inception_4d_5x5_reduce = Conv2D(32, (1,1), padding='same', activation='relu', name='inception_4d/5x5_reduce', kernel_regularizer=l2(0.0002))(inception_4c_output)
  110. inception_4d_5x5_pad = ZeroPadding2D(padding=(2, 2))(inception_4d_5x5_reduce)
  111. inception_4d_5x5 = Conv2D(64, (5,5), padding='valid', activation='relu', name='inception_4d/5x5', kernel_regularizer=l2(0.0002))(inception_4d_5x5_pad)
  112. inception_4d_pool = MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same', name='inception_4d/pool')(inception_4c_output)
  113. inception_4d_pool_proj = Conv2D(64, (1,1), padding='same', activation='relu', name='inception_4d/pool_proj', kernel_regularizer=l2(0.0002))(inception_4d_pool)
  114. inception_4d_output = Concatenate(axis=1, name='inception_4d/output')([inception_4d_1x1,inception_4d_3x3,inception_4d_5x5,inception_4d_pool_proj])
  115.  
  116. loss2_ave_pool = AveragePooling2D(pool_size=(5,5), strides=(3,3), name='loss2/ave_pool')(inception_4d_output)
  117. loss2_conv = Conv2D(128, (1,1), padding='same', activation='relu', name='loss2/conv', kernel_regularizer=l2(0.0002))(loss2_ave_pool)
  118. loss2_flat = Flatten()(loss2_conv)
  119. loss2_fc = Dense(1024, activation='relu', name='loss2/fc', kernel_regularizer=l2(0.0002))(loss2_flat)
  120. loss2_drop_fc = Dropout(rate=0.7)(loss2_fc)
  121. loss2_classifier = Dense(1000, name='loss2/classifier', kernel_regularizer=l2(0.0002))(loss2_drop_fc)
  122. loss2_classifier_act = Activation('softmax')(loss2_classifier)
  123.  
  124. inception_4e_1x1 = Conv2D(256, (1,1), padding='same', activation='relu', name='inception_4e/1x1', kernel_regularizer=l2(0.0002))(inception_4d_output)
  125. inception_4e_3x3_reduce = Conv2D(160, (1,1), padding='same', activation='relu', name='inception_4e/3x3_reduce', kernel_regularizer=l2(0.0002))(inception_4d_output)
  126. inception_4e_3x3_pad = ZeroPadding2D(padding=(1, 1))(inception_4e_3x3_reduce)
  127. inception_4e_3x3 = Conv2D(320, (3,3), padding='valid', activation='relu', name='inception_4e/3x3', kernel_regularizer=l2(0.0002))(inception_4e_3x3_pad)
  128. inception_4e_5x5_reduce = Conv2D(32, (1,1), padding='same', activation='relu', name='inception_4e/5x5_reduce', kernel_regularizer=l2(0.0002))(inception_4d_output)
  129. inception_4e_5x5_pad = ZeroPadding2D(padding=(2, 2))(inception_4e_5x5_reduce)
  130. inception_4e_5x5 = Conv2D(128, (5,5), padding='valid', activation='relu', name='inception_4e/5x5', kernel_regularizer=l2(0.0002))(inception_4e_5x5_pad)
  131. inception_4e_pool = MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same', name='inception_4e/pool')(inception_4d_output)
  132. inception_4e_pool_proj = Conv2D(128, (1,1), padding='same', activation='relu', name='inception_4e/pool_proj', kernel_regularizer=l2(0.0002))(inception_4e_pool)
  133. inception_4e_output = Concatenate(axis=1, name='inception_4e/output')([inception_4e_1x1,inception_4e_3x3,inception_4e_5x5,inception_4e_pool_proj])
  134.  
  135. inception_4e_output_zero_pad = ZeroPadding2D(padding=(1, 1))(inception_4e_output)
  136. pool4_helper = PoolHelper()(inception_4e_output_zero_pad)
  137. pool4_3x3_s2 = MaxPooling2D(pool_size=(3,3), strides=(2,2), padding='valid', name='pool4/3x3_s2')(pool4_helper)
  138.  
  139. inception_5a_1x1 = Conv2D(256, (1,1), padding='same', activation='relu', name='inception_5a/1x1', kernel_regularizer=l2(0.0002))(pool4_3x3_s2)
  140. inception_5a_3x3_reduce = Conv2D(160, (1,1), padding='same', activation='relu', name='inception_5a/3x3_reduce', kernel_regularizer=l2(0.0002))(pool4_3x3_s2)
  141. inception_5a_3x3_pad = ZeroPadding2D(padding=(1, 1))(inception_5a_3x3_reduce)
  142. inception_5a_3x3 = Conv2D(320, (3,3), padding='valid', activation='relu', name='inception_5a/3x3', kernel_regularizer=l2(0.0002))(inception_5a_3x3_pad)
  143. inception_5a_5x5_reduce = Conv2D(32, (1,1), padding='same', activation='relu', name='inception_5a/5x5_reduce', kernel_regularizer=l2(0.0002))(pool4_3x3_s2)
  144. inception_5a_5x5_pad = ZeroPadding2D(padding=(2, 2))(inception_5a_5x5_reduce)
  145. inception_5a_5x5 = Conv2D(128, (5,5), padding='valid', activation='relu', name='inception_5a/5x5', kernel_regularizer=l2(0.0002))(inception_5a_5x5_pad)
  146. inception_5a_pool = MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same', name='inception_5a/pool')(pool4_3x3_s2)
  147. inception_5a_pool_proj = Conv2D(128, (1,1), padding='same', activation='relu', name='inception_5a/pool_proj', kernel_regularizer=l2(0.0002))(inception_5a_pool)
  148. inception_5a_output = Concatenate(axis=1, name='inception_5a/output')([inception_5a_1x1,inception_5a_3x3,inception_5a_5x5,inception_5a_pool_proj])
  149.  
  150. inception_5b_1x1 = Conv2D(384, (1,1), padding='same', activation='relu', name='inception_5b/1x1', kernel_regularizer=l2(0.0002))(inception_5a_output)
  151. inception_5b_3x3_reduce = Conv2D(192, (1,1), padding='same', activation='relu', name='inception_5b/3x3_reduce', kernel_regularizer=l2(0.0002))(inception_5a_output)
  152. inception_5b_3x3_pad = ZeroPadding2D(padding=(1, 1))(inception_5b_3x3_reduce)
  153. inception_5b_3x3 = Conv2D(384, (3,3), padding='valid', activation='relu', name='inception_5b/3x3', kernel_regularizer=l2(0.0002))(inception_5b_3x3_pad)
  154. inception_5b_5x5_reduce = Conv2D(48, (1,1), padding='same', activation='relu', name='inception_5b/5x5_reduce', kernel_regularizer=l2(0.0002))(inception_5a_output)
  155. inception_5b_5x5_pad = ZeroPadding2D(padding=(2, 2))(inception_5b_5x5_reduce)
  156. inception_5b_5x5 = Conv2D(128, (5,5), padding='valid', activation='relu', name='inception_5b/5x5', kernel_regularizer=l2(0.0002))(inception_5b_5x5_pad)
  157. inception_5b_pool = MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same', name='inception_5b/pool')(inception_5a_output)
  158. inception_5b_pool_proj = Conv2D(128, (1,1), padding='same', activation='relu', name='inception_5b/pool_proj', kernel_regularizer=l2(0.0002))(inception_5b_pool)
  159. inception_5b_output = Concatenate(axis=1, name='inception_5b/output')([inception_5b_1x1,inception_5b_3x3,inception_5b_5x5,inception_5b_pool_proj])
  160.  
  161. pool5_7x7_s1 = AveragePooling2D(pool_size=(7,7), strides=(1,1), name='pool5/7x7_s2')(inception_5b_output)
  162. loss3_flat = Flatten()(pool5_7x7_s1)
  163. pool5_drop_7x7_s1 = Dropout(rate=0.4)(loss3_flat)
  164. loss3_classifier = Dense(1000, name='loss3/classifier', kernel_regularizer=l2(0.0002))(pool5_drop_7x7_s1)
  165. loss3_classifier_act = Activation('softmax', name='prob')(loss3_classifier)
  166.  
  167. googlenet = Model(inputs=input, outputs=[loss1_classifier_act,loss2_classifier_act,loss3_classifier_act])
  168.  
  169. if weights_path:
  170. googlenet.load_weights(weights_path)
  171.  
  172. if keras.backend.backend() == 'tensorflow':
  173. # convert the convolutional kernels for tensorflow
  174. ops = []
  175. for layer in googlenet.layers:
  176. if layer.__class__.__name__ == 'Conv2D':
  177. original_w = K.get_value(layer.kernel)
  178. converted_w = convert_kernel(original_w)
  179. ops.append(tf.assign(layer.kernel, converted_w).op)
  180. K.get_session().run(ops)
  181.  
  182. return googlenet
  183.  
  184.  
  185. if __name__ == "__main__":
  186. img = imageio.imread('cat.jpg', pilmode='RGB')
  187. img = np.array(Image.fromarray(img).resize((224, 224))).astype(np.float32)
  188. img[:, :, 0] -= 123.68
  189. img[:, :, 1] -= 116.779
  190. img[:, :, 2] -= 103.939
  191. img[:,:,[0,1,2]] = img[:,:,[2,1,0]]
  192. img = img.transpose((2, 0, 1))
  193. img = np.expand_dims(img, axis=0)
  194.  
  195. # Test pretrained model
  196. model = create_googlenet('googlenet_weights.h5')
  197. sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
  198. model.compile(optimizer=sgd, loss='categorical_crossentropy')
  199. out = model.predict(img) # note: the model has three outputs
  200. labels = np.loadtxt('synset_words.txt', str, delimiter='\t')
  201. predicted_label = np.argmax(out[2])
  202. predicted_class_name = labels[predicted_label]
  203. print('Predicted Class: ', predicted_label, ', Class Name: ', predicted_class_name)
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!