create_model_maggiori17_fullyconv.py 10 KB
Newer Older
remi cresson's avatar
remi cresson committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

# Imports
import numpy as np
import tensorflow as tf
import numpy as np
import random
import os
import shutil
import sys
import time
from sklearn.metrics import accuracy_score
from sklearn.metrics import precision_recall_fscore_support
from sklearn.metrics import f1_score
from sklearn.ensemble import RandomForestClassifier
from sklearn.utils import shuffle
from sklearn.metrics import confusion_matrix

from tricks import *

tf.logging.set_verbosity(tf.logging.INFO)

def do_eval2(sess,
            testPrediction,
            xs_placeholder,
            labels_placeholder,
            istraining_placeholder,
            ds_data_xs,
            ds_labels,
            batch_size):
  """Runs one evaluation against the full epoch of data.
  """
  # total predictions
  tot_pred = np.array([])
  
  # And run one epoch of eval.
  n_data = ds_data_xs.shape[0]
  n_steps = int(n_data / batch_size)
  for step in range(0, n_steps):
    start_idx = batch_size * step
    end_idx = start_idx + batch_size
    if step == n_steps - 1:
        end_idx = n_data
        print(end_idx)

    feed_dict = {
        xs_placeholder: ds_data_xs[start_idx:end_idx,:],
        labels_placeholder: ds_labels[start_idx:end_idx],
        istraining_placeholder: False,
    }
    ( predicted_y) = sess.run(testPrediction, feed_dict=feed_dict)
    
    if tot_pred.size == 0:
      tot_pred = predicted_y
    else:
      tot_pred = np.concatenate( (tot_pred, predicted_y), axis=0)

  print(ds_labels.shape)
  print(tot_pred.shape)
    
  tot_pred = tot_pred.reshape((tot_pred.size))
  ds_labels = ds_labels.reshape((ds_labels.size))
  print(tot_pred.shape)

  print("F-Measure: "+str(f1_score(ds_labels, tot_pred, average='weighted')))
  print("F1 score: " + str(f1_score(ds_labels, tot_pred, average=None)))
  print("Accuracy: "+str(accuracy_score(ds_labels, tot_pred)))

def main(unused_argv):
  """ Main function
  In this function we do:
    1. Import a dataset
    2. Build a model implementing a CNN
    3. Perform training of the CNN
    4. Export the model
  """

  # Training params
  n_epoch = 100
  batch_size = 32
  learning_rate = 0.0001

  ############################################################
  #                     import a dataset 
  ############################################################


  # check number of arguments
  if len(sys.argv) != 4:
    print("Usage : <patches> <labels> <output_model_dir>")
    sys.exit(1)

  # Export dir
  log_dir = sys.argv[3] + '/model_checkpoints/'
  export_dir = sys.argv[3] + '/model_export/'

  print("loading dataset")

  # Create a dataset of size imp_size
  imp_ds_patches = read_samples(sys.argv[1])
  imp_ds_labels  = read_samples(sys.argv[2])
  
  # Shuffle the dataset
  imp_ds_patches,imp_ds_labels = shuffle(imp_ds_patches,imp_ds_labels, random_state=0)
  
  print("ok")
  
   # Number of samples
  if (imp_ds_patches.shape[0] != imp_ds_labels.shape[0]):
    print("Number of samples should be the same as number of patches!")
    sys.exit(1)

  # Number of samples for training  
  n_data_train = int(imp_ds_patches.shape[0] / 2)

  # Size of patches  
  nb_bands = 4
  patch_size_xs = 80
  patch_size_label = 16
  
  # Divide the dataset in two subdatasets : training and validation
  ds_data_train = imp_ds_patches[0:n_data_train,:]
  ds_data_valid = imp_ds_patches[n_data_train:imp_ds_patches.shape[0],:]
  ds_labels_train = imp_ds_labels[0:n_data_train,:]
  ds_labels_valid = imp_ds_labels[n_data_train:imp_ds_patches.shape[0],:]

  ############################################################
  #                    Build the graph
  ############################################################
  
  with tf.Graph().as_default():

    # placeholder for images and labels
    istraining_placeholder = tf.placeholder(tf.bool, shape=(), name="istraining") # Used only for dropout...
    xs_placeholder = tf.placeholder(tf.float32, shape=(None, patch_size_xs, patch_size_xs, nb_bands), name="x1")
    labels_placeholder = tf.placeholder(tf.int32, shape=(None, patch_size_label, patch_size_label, 1), name="y1")
 
    print_tensor_info("xs_placeholder",xs_placeholder)
 
    # Convolutional Layer #1 
    conv1 = tf.layers.conv2d(
      inputs=xs_placeholder,
      filters=32,
      kernel_size=[12, 12],
      padding="valid",
      activation=tf.nn.crelu)
    print_tensor_info("conv1",conv1)
  
    # Normalization of output of layer 1
    norm1 = tf.layers.batch_normalization(conv1)
    print_tensor_info("norm1",norm1)
  
    # pooling layer #1
    pool1 = tf.layers.max_pooling2d(inputs=norm1, pool_size=[4, 4], strides=4)
    print_tensor_info("pool1",pool1)

    # Convolutional Layer #2
    conv2 = tf.layers.conv2d(
      inputs=pool1,
      filters=56,
      kernel_size=[4, 4],
      padding="valid",
      activation=tf.nn.crelu)
    print_tensor_info("conv2",conv2)

     # Normalization of output of layer 2
    norm2 = tf.layers.batch_normalization(conv2)
    print_tensor_info("norm2",norm2)
   
    # Convolutional Layer #3 
    conv3 = tf.layers.conv2d(
      inputs=norm2,
      filters=40,
      kernel_size=[3, 3],
      padding="valid",
      activation=tf.nn.crelu)
    print_tensor_info("conv3",conv3)
 
     # Normalization of output of layer 3
    norm3 = tf.layers.batch_normalization(conv3)
    print_tensor_info("norm3",norm3)
  
    # Convolutional Layer #4
    conv4 = tf.layers.conv2d(
      inputs=norm3,
      filters=1,
      kernel_size=[8, 8],
      padding="valid",
      activation=tf.nn.crelu)
    print_tensor_info("conv4",conv4)
    
    # Deconv = conv on the padded/strided input, that is an (5+1)*4
    deconv1 = tf.layers.conv2d_transpose(
      inputs=conv4,
      filters=1,
      strides=(4,4),
      kernel_size=[8, 8],
      padding="valid",
      activation=tf.nn.sigmoid)
    print_tensor_info("deconv1",deconv1)
 
    numbatch = tf.shape(deconv1)[0]
    szx =      tf.shape(deconv1)[1]
    szy =      tf.shape(deconv1)[2]
    estimated = tf.slice(deconv1, [0, 4, 4, 0], [numbatch, szx - 8, szy - 8, 1], "estimated")
    print_tensor_info("estimated", estimated)
   
    ##################### Loss calculation ###################

    estimated_resized = tf.reshape(estimated, [-1, patch_size_label*patch_size_label])
    labels_resized = tf.reshape(labels_placeholder, [-1, patch_size_label*patch_size_label])
    labels_resized = tf.cast(labels_resized, tf.float32)
    
    testPrediction = tf.to_int32(estimated > 0.5)

    loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels=labels_resized, logits=estimated_resized))
    
    ###################### Gradient descent ##################
    
    train_op = tf.train.AdamOptimizer(learning_rate, name="optimizer").minimize(loss)
    
    ############### Variable initializer Op ##################

    init = tf.global_variables_initializer()
    
    ######################### Saver ##########################

    saver = tf.train.Saver()
    
    #################### Create a session ####################
    
    sess = tf.Session()
    sess.run(init)

    ############## Here we start the training ################
    
    for curr_epoch in range(n_epoch):

      print("Epoch #" + str(curr_epoch))
      
      ds_data_train,ds_labels_train = shuffle(ds_data_train,ds_labels_train, random_state=0)
  
      # Start the training loop.
      n_steps = int(n_data_train / batch_size)
      for step in range(n_steps):
  
        start_time = time.time()
        
        # Fill a feed dictionary with the actual set of images and labels
        # for this particular training step.
        # Load training and eval data
        start_idx = batch_size * step
        end_idx = start_idx + batch_size 
        feed_dict = {
          xs_placeholder: ds_data_train[start_idx:end_idx,:],
          labels_placeholder: ds_labels_train[start_idx:end_idx,:],
          istraining_placeholder: True,
        }
  
        # Run one step of the model.  The return values are the activations
        # from the `train_op` (which is discarded) and the `loss` Op.  To
        # inspect the values of your Ops or variables, you may include them
        # in the list passed to sess.run() and the value tensors will be
        # returned in the tuple from the call.
        _, loss_value = sess.run([train_op, loss], feed_dict=feed_dict)

        duration = time.time() - start_time
  
        # Print an overview fairly often.
        if step % 10 == 0:
          # Print status to stdout.
          print('Step %d: loss = %.2f (%.3f sec)' % (step, loss_value, duration))
          #print('Step %d: (%.3f sec)' % (step, duration))


      # Save a checkpoint and evaluate the model periodically.
      if (curr_epoch + 1) % 1 == 0:
        checkpoint_file = os.path.join(log_dir, 'model.ckpt')
        saver.save(sess, checkpoint_file, global_step=step)
        # Evaluate against the training set.
        print('Training Data Eval:')
        do_eval2(sess,
                testPrediction,
                xs_placeholder,
                labels_placeholder,
                istraining_placeholder,
                ds_data_train,
                ds_labels_train,
                batch_size)
        # Evaluate against the validation set.
        print('Validation Data Eval:')
        do_eval2(sess,
                testPrediction,
                xs_placeholder,
                labels_placeholder,
                istraining_placeholder,
                ds_data_valid,
                ds_labels_valid,
                batch_size)
                
        # Let's export a SavedModel
        shutil.rmtree(export_dir)
        builder = tf.saved_model.builder.SavedModelBuilder(export_dir)
        signature_def_map= {
        "model": tf.saved_model.signature_def_utils.predict_signature_def(
            inputs= {"x1": xs_placeholder},
            outputs= {"prediction": testPrediction})
        }
        builder.add_meta_graph_and_variables(sess, [tf.saved_model.tag_constants.TRAINING], signature_def_map)
        builder.add_meta_graph([tf.saved_model.tag_constants.SERVING])
        builder.save()

  quit()
  
if __name__ == "__main__":
  
  tf.add_check_numerics_ops()
  tf.app.run(main)