手写汉字笔迹识别模型汇总

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
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
手写汉字笔迹识别模型：
第一名用的是googleNet，准确率97.3%
def GoogleLeNetSlim(x, num_classes, keep_prob=0.5):
    with tf.variable_scope('main'):
        t = slim.conv2d(x, 64, [3, 3], [1, 1], padding='SAME', activation_fn=relu, normalizer_fn=slim.batch_norm, scope='conv1')
        t = slim.max_pool2d(t, [2, 2], [2, 2], padding='SAME')
        t = slim.conv2d(t, 96, [3, 3], [1, 1], padding='SAME', activation_fn=relu, normalizer_fn=slim.batch_norm, scope='conv2')
        t = slim.conv2d(t, 192, [3, 3], [1, 1], padding='SAME', activation_fn=relu, normalizer_fn=slim.batch_norm, scope='conv3')
        t = slim.max_pool2d(t, [2, 2], [2, 2], padding='SAME')
 
    with tf.variable_scope('block1'):
        t = block_slim(t, [64, 96, 128, 16, 32, 32], name='block1')       # [?, 16, 16, 256]
 
    with tf.variable_scope('block2'):
        t = block_slim(t, [128, 128, 192, 32, 96, 64], name='block1')     # [?, 16, 16, 480]
        t = tf.nn.max_pool(t, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
 
    with tf.variable_scope('block3'):
        t = block_slim(t, [192, 96, 208, 16, 48, 64], name='block1')
        t = block_slim(t, [160, 112, 224, 24, 64, 64], name='block2')
        t = block_slim(t, [128, 128, 256, 24, 64, 64], name='block3')
        t = block_slim(t, [112, 144, 288, 32, 64, 64], name='block4')
        t = block_slim(t, [256, 160, 320, 32, 128, 128], name='block5')    # [?, 8, 8, 832]
        t = tf.nn.max_pool(t, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
 
    with tf.variable_scope('block4'):
        t = block_slim(t, [256, 160, 320, 32, 128, 128], name='block1')
        t = block_slim(t, [384, 192, 384, 48, 128, 128], name='block2')    # [?, 8, 8, 1024]
        t = tf.nn.max_pool(t, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
 
    with tf.variable_scope('fc'):
        t = slim.flatten(t)
        t = slim.fully_connected(slim.dropout(t, keep_prob), 1024, activation_fn=relu, normalizer_fn=slim.batch_norm, scope='fc1')
        t = slim.fully_connected(slim.dropout(t, keep_prob), num_classes, activation_fn=None, scope='logits')
 
    return t
TODO：实验下，https://github.com/tflearn/tflearn/blob/master/examples/images/googlenet.py
 
 
还有使用inception v3的！！！
def build_graph_all(top_k,scope=None):
    keep_prob = tf.placeholder(dtype=tf.float32, shape=[], name='keep_prob')
    images = tf.placeholder(dtype=tf.float32, shape=[None, image_size, image_size, 1], name='image_batch')
    labels = tf.placeholder(dtype=tf.int64, shape=[None], name='label_batch')
 
    with tf.variable_scope(scope,'Incept_Net',[images]):
        with slim.arg_scope([slim.conv2d,slim.max_pool2d,slim.avg_pool2d],stride=1,padding='VALID'):
 
            net = slim.conv2d(images,32,[3,3],scope='conv2d_1a_3x3')
            print('tensor 1:' + str(net.get_shape().as_list()))
 
            net = slim.conv2d(net,32,[3,3],scope='conv2d_2a_3x3')
            print('tensor 2:' + str(net.get_shape().as_list()))
 
            net = slim.conv2d(net,64,[3,3],padding='SAME',scope='conv2d_2b_3x3')
            print('tensor 3:' + str(net.get_shape().as_list()))
 
            net = slim.max_pool2d(net,[3,3],stride=2,scope='maxpool_3a_3x3')
            print('tensor 4:' + str(net.get_shape().as_list()))
 
            net = slim.conv2d(net,80,[1,1],scope='conv2d_3b_1x1')
            print('tensor 5:' + str(net.get_shape().as_list()))
 
            net = slim.conv2d(net,192,[3,3],scope='conv2d_4a_3x3')
            print('tensor 6:' + str(net.get_shape().as_list()))
 
            net = slim.max_pool2d(net,[3,3],stride=2,scope='maxpool_5a_3x3')
            print('tensor 7:' + str(net.get_shape().as_list()))
 
        with slim.arg_scope([slim.conv2d,slim.max_pool2d,slim.avg_pool2d],stride=1,padding='SAME'):
            with tf.variable_scope('mixed_5b'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,64,[1,1],scope='conv2d_0a_1x1')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,48,[1,1],scope='conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1,64,[5,5],scope='conv2d_0b_5x5')
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.conv2d(net,64,[1,1],scope='conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2,96,[3,3],scope='conv2d_0b_3x3')
                    branch_2 = slim.conv2d(branch_2,96,[3,3],scope='conv2d_0c_3x3')
                with tf.variable_scope('branch_3'):
                    branch_3 = slim.avg_pool2d(net,[3,3],scope='avgpool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3,32,[1,1],scope='conv2d_0b_1x1')
 
                net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)
                print('tensor 8:' + str(net.get_shape().as_list()))
 
 
            with tf.variable_scope('mixed_5c'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,64,[1,1],scope='conv2d_0a_1x1')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,48,[1,1],scope='conv2d_0b_1x1')
                    branch_1 = slim.conv2d(branch_1,64,[5,5],scope='conv2d_0c_5x5')
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.conv2d(net,64,[1,1],scope='conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2,96,[3,3],scope='conv2d_0b_3x3')
                    branch_2 = slim.conv2d(branch_2,96,[3,3],scope='conv2d_0c_3x3')
                with tf.variable_scope('branch_3'):
                    branch_3 = slim.avg_pool2d(net,[3,3],scope='avgpool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3,64,[1,1],scope='conv2d_0b_1x1')
 
                net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)
                print('tensor 9:' + str(net.get_shape().as_list()))
 
 
            with tf.variable_scope('mixed_5d'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,64,[1,1],scope='conv2d_0a_1x1')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,48,[1,1],scope='conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1,64,[5,5],scope='conv2d_0b_5x5')
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.conv2d(net,64,[1,1],scope='conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2,96,[3,3],scope='conv2d_0b_3x3')
                    branch_2 = slim.conv2d(branch_2,96,[3,3],scope='conv2d_0c_3x3')
                with tf.variable_scope('branch_3'):
                    branch_3 = slim.avg_pool2d(net,[3,3],scope='avgpool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3,64,[1,1],scope='conv2d_0b_1x1')
 
                net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)
                print('tensor 10:' + str(net.get_shape().as_list()))
 
 
            with tf.variable_scope('mixed_6a'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,384,[3,3],stride=2,padding='VALID',scope='conv2d_1a_1x1')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,64,[1,1],scope='conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1,96,[3,3],scope='conv2d_0b_3x3')
                    branch_1 = slim.conv2d(branch_1,96,[3,3],stride=2,padding='VALID',scope='conv2d_1a_1x1')
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.max_pool2d(net,[3,3],stride=2,padding='VALID',scope='maxpool_1a_3x3')
 
                net = tf.concat([branch_0,branch_1,branch_2],3)
                print('tensor 11:' + str(net.get_shape().as_list()))
 
 
            with tf.variable_scope('mixed_6b'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,192,[1,1],scope='conv2d_0a_1x1')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,128,[1,1],scope='conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1,128,[1,7],scope='conv2d_0b_1x7')
                    branch_1 = slim.conv2d(branch_1,192,[7,1],scope='conv2d_0c_7x1')
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.conv2d(net,128,[1,1],scope='conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2,128,[7,1],scope='conv2d_0b_7x1')
                    branch_2 = slim.conv2d(branch_2,128,[1,7],scope='conv2d_0c_1x7')
                    branch_2 = slim.conv2d(branch_2,128,[7,1],scope='conv2d_0d_7x1')
                    branch_2 = slim.conv2d(branch_2,192,[1,7],scope='conv2d_0e_1x7')
                with tf.variable_scope('branch_3'):
                    branch_3 = slim.avg_pool2d(net,[3,3],scope='avgpool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3,192,[1,1],scope='conv2d_0b_1x1')
 
                net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)
                print('tensor 12:' + str(net.get_shape().as_list()))
 
 
            with tf.variable_scope('mixed_6c'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,192,[1,1],scope='conv2d_0a_1x1')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,160,[1,1],scope='conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1,160,[1,7],scope='conv2d_0b_1x7')
                    branch_1 = slim.conv2d(branch_1,192,[7,1],scope='conv2d_0c_7x1')
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.conv2d(net,160,[1,1],scope='conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2,160,[7,1],scope='conv2d_0b_7x1')
                    branch_2 = slim.conv2d(branch_2,160,[1,7],scope='conv2d_0c_1x7')
                    branch_2 = slim.conv2d(branch_2,160,[7,1],scope='conv2d_0d_7x1')
                    branch_2 = slim.conv2d(branch_2,192,[1,7],scope='conv2d_0e_1x7')
                with tf.variable_scope('branch_3'):
                    branch_3 = slim.avg_pool2d(net,[3,3],scope='avgpool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3,192,[1,1],scope='conv2d_0b_1x1')
 
                net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)
                print('tensor 13:' + str(net.get_shape().as_list()))
 
 
            with tf.variable_scope('mixed_6d'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,192,[1,1],scope='conv2d_0a_1x1')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,160,[1,1],scope='conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1,160,[1,7],scope='conv2d_0b_1x7')
                    branch_1 = slim.conv2d(branch_1,192,[7,1],scope='conv2d_0c_7x1')
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.conv2d(net,160,[1,1],scope='conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2,160,[7,1],scope='conv2d_0b_7x1')
                    branch_2 = slim.conv2d(branch_2,160,[1,7],scope='conv2d_0c_1x7')
                    branch_2 = slim.conv2d(branch_2,160,[7,1],scope='conv2d_0d_7x1')
                    branch_2 = slim.conv2d(branch_2,192,[1,7],scope='conv2d_0e_1x7')
                with tf.variable_scope('branch_3'):
                    branch_3 = slim.avg_pool2d(net,[3,3],scope='avgpool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3,192,[1,1],scope='conv2d_0b_1x1')
 
                net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)
                print('tensor 14:' + str(net.get_shape().as_list()))
 
 
            with tf.variable_scope('mixed_6e'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,192,[1,1],scope='conv2d_0a_1x1')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,192,[1,1],scope='conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1,192,[1,7],scope='conv2d_0b_1x7')
                    branch_1 = slim.conv2d(branch_1,192,[7,1],scope='conv2d_0c_7x1')
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.conv2d(net,192,[1,1],scope='conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2,192,[7,1],scope='conv2d_0b_7x1')
                    branch_2 = slim.conv2d(branch_2,192,[1,7],scope='conv2d_0c_1x7')
                    branch_2 = slim.conv2d(branch_2,192,[7,1],scope='conv2d_0d_7x1')
                    branch_2 = slim.conv2d(branch_2,192,[1,7],scope='conv2d_0e_1x7')
                with tf.variable_scope('branch_3'):
                    branch_3 = slim.avg_pool2d(net,[3,3],scope='avgpool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3,192,[1,1],scope='conv2d_0b_1x1')
 
                net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)
                print('tensor 15:' + str(net.get_shape().as_list()))
 
 
            with tf.variable_scope('mixed_7a'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,192,[1,1],scope='conv2d_0a_1x1')
                    branch_0 = slim.conv2d(branch_0,320,[3,3],stride=2,padding='VALID',scope='conv2d_1a_3x3')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,192,[1,1],scope='conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1,192,[1,7],scope='conv2d_0b_1x7')
                    branch_1 = slim.conv2d(branch_1,192,[7,1],scope='conv2d_0c_7x1')
                    branch_1 = slim.conv2d(branch_1,192,[3,3],stride=2,padding='VALID',scope='conv2d_1a_3x3')
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.max_pool2d(net,[3,3],stride=2,padding='VALID',scope='maxpool_1a_3x3')
 
                net = tf.concat([branch_0,branch_1,branch_2],3)
                print('tensor 16:' + str(net.get_shape().as_list()))
 
 
            with tf.variable_scope('mixed_7b'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,320,[1,1],scope='conv2d_0a_1x1')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,384,[1,1],scope='conv2d_0a_1x1')
                    branch_1 = tf.concat([
                        slim.conv2d(branch_1,384,[1,3],scope='conv2d_0b_1x3'),
                        slim.conv2d(branch_1,384,[3,1],scope='conv2d_0b_3x1')
                    ],3)
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.conv2d(net,448,[1,1],scope='conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2,384,[3,3],scope='conv2d_0b_3x3')
                    branch_2 = tf.concat([
                        slim.conv2d(branch_2,384,[1,3],scope='conv2d_0c_1x3'),
                        slim.conv2d(branch_2,384,[3,1],scope='conv2d_0d_3x1')
                    ],3)
                with tf.variable_scope('branch_3'):
                    branch_3 = slim.avg_pool2d(net,[3,3],scope='avgpool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3,192,[1,1],scope='conv2d_0b_1x1')
 
                net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)
                print('tensor 17:' + str(net.get_shape().as_list()))
 
 
            with tf.variable_scope('mixed_7c'):
                with tf.variable_scope('branch_0'):
                    branch_0 = slim.conv2d(net,320,[1,1],scope='conv2d_0a_1x1')
                with tf.variable_scope('branch_1'):
                    branch_1 = slim.conv2d(net,384,[1,1],scope='conv2d_0a_1x1')
                    branch_1 = tf.concat([
                        slim.conv2d(branch_1,384,[1,3],scope='conv2d_0b_1x3'),
                        slim.conv2d(branch_1,384,[3,1],scope='conv2d_0c_3x1')],3)
                with tf.variable_scope('branch_2'):
                    branch_2 = slim.conv2d(net,448,[1,1],scope='conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2,384,[3,3],scope='conv2d_0b_3x3')
                    branch_2 = tf.concat([
                        slim.conv2d(branch_2,384,[1,3],scope='conv2d_0c_1x3'),
                        slim.conv2d(branch_2,384,[3,1],scope='conv2d_0d_3x1')],3)
                with tf.variable_scope('branch_3'):
                    branch_3 = slim.avg_pool2d(net,[3,3],scope='avgpool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3,192,[1,1],scope='conv2d_0b_1x1')
 
                net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)
                print('tensor 18:' + str(net.get_shape().as_list()))
 
    with slim.arg_scope([slim.conv2d,slim.max_pool2d,slim.avg_pool2d],stride=1,padding='SAME'):
        with tf.variable_scope('logits'):
            net = slim.avg_pool2d(net,[3,3],padding='VALID',scope='avgpool_1a_3x3')
            print('tensor 19:' + str(net.get_shape().as_list()))
 
            net = slim.dropout(net,keep_prob=keep_prob,scope='dropout_1b')
 
            logits = slim.conv2d(net, char_size,[2,2],padding='VALID',activation_fn=None,normalizer_fn=None,
                                 scope='conv2d_1c_2x2')
            print('logits 1:' + str(logits.get_shape().as_list()))
 
            logits = tf.squeeze(logits,[1,2],name='spatialsqueeze')
            print('logits 2:' + str(logits.get_shape().as_list()))
 
    regularization_loss = tf.reduce_sum(tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES))
    loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits, labels=labels))
 
    total_loss = loss + regularization_loss
    print('get total_loss')
 
    accuracy = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(logits, 1), labels), tf.float32))
 
    global_step = tf.get_variable("step", [], initializer=tf.constant_initializer(0.0), trainable=False)
    rate = tf.train.exponential_decay(2e-3, global_step, decay_steps=2000, decay_rate=0.97, staircase=True)
 
    update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
    with tf.control_dependencies(update_ops):
        train_op = tf.train.AdamOptimizer(learning_rate=rate).minimize(total_loss, global_step=global_step)
 
    probabilities = tf.nn.softmax(logits)
 
    tf.summary.scalar('loss', loss)
    tf.summary.scalar('accuracy', accuracy)
    merged_summary_op = tf.summary.merge_all()
    predicted_val_top_k, predicted_index_top_k = tf.nn.top_k(probabilities, k=top_k)
    accuracy_in_top_k = tf.reduce_mean(tf.cast(tf.nn.in_top_k(probabilities, labels, top_k), tf.float32))
 
    return {'images': images,
            'labels': labels,
            'keep_prob': keep_prob,
            'top_k': top_k,
            'global_step': global_step,
            'train_op': train_op,
            'loss': total_loss,
            'accuracy': accuracy,
            'accuracy_top_k': accuracy_in_top_k,
            'merged_summary_op': merged_summary_op,
            'predicted_distribution': probabilities,
            'predicted_index_top_k': predicted_index_top_k,
            'predicted_val_top_k': predicted_val_top_k}
 
用resnet v2的:
resnet_v2.default_image_size = 128
 
 
def resnet_v2_50(inputs,
                 num_classes=None,
                 is_training=True,
                 global_pool=True,
                 output_stride=None,
                 spatial_squeeze=True,
                 reuse=None,
                 scope='resnet_v2_50'):
    """ResNet-50 model of [1]. See resnet_v2() for arg and return description."""
    blocks = [
        resnet_v2_block('block1', base_depth=64, num_units=3, stride=2),
        resnet_v2_block('block2', base_depth=128, num_units=4, stride=2),
        resnet_v2_block('block3', base_depth=256, num_units=6, stride=2),
        resnet_v2_block('block4', base_depth=512, num_units=3, stride=1),
    ]
    return resnet_v2(inputs, blocks, num_classes, is_training=is_training,
                     global_pool=global_pool, output_stride=output_stride,
                     include_root_block=True, spatial_squeeze=spatial_squeeze,
                     reuse=reuse, scope=scope)
 
 
resnet_v2_50.default_image_size = resnet_v2.default_image_size
 
 
def resnet_v2_101(inputs,
                  num_classes=None,
                  is_training=True,
                  global_pool=True,
                  output_stride=None,
                  spatial_squeeze=True,
                  reuse=None,
                  scope='resnet_v2_101'):
    """ResNet-101 model of [1]. See resnet_v2() for arg and return description."""
    blocks = [
        resnet_v2_block('block1', base_depth=64, num_units=3, stride=2),
        resnet_v2_block('block2', base_depth=128, num_units=4, stride=2),
        resnet_v2_block('block3', base_depth=256, num_units=23, stride=2),
        resnet_v2_block('block4', base_depth=512, num_units=3, stride=1),
    ]
    return resnet_v2(inputs, blocks, num_classes, is_training=is_training,
                     global_pool=global_pool, output_stride=output_stride,
                     include_root_block=True, spatial_squeeze=spatial_squeeze,
                     reuse=reuse, scope=scope)
                      
def build_graph(top_k, is_training):
    # with tf.device('/cpu:0'):
    keep_prob = tf.placeholder(dtype=tf.float32, shape=[], name='keep_prob')
    images = tf.placeholder(dtype=tf.float32, shape=[None, 128, 128, 1], name='image_batch')
    labels = tf.placeholder(dtype=tf.int64, shape=[None], name='label_batch')
 
    logits, _ = resnet_v2_50(images, num_classes=3755, is_training=is_training, global_pool=True,
                             output_stride=None, spatial_squeeze=True, reuse=None)  
     
-----------------------------  