[38391a]: / he_j_inference / keras_model.py

Download this file

429 lines (398 with data), 20.1 kB

  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
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
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
#coding:utf-8
import importlib
import keras.backend as K
from keras.engine import InputSpec
from keras.layers import Input,Lambda,Dropout,Concatenate
from keras.activations import softmax
from keras.layers.core import Dense
from keras.layers import Conv2D,Average,MaxPooling2D,AveragePooling2D,Add,Flatten
from keras.layers import GlobalMaxPooling2D,GlobalAveragePooling2D,Multiply,LocallyConnected2D
from keras.models import Model
#import cv2
from keras.engine.topology import Layer
import numpy as np
import tensorflow as tf
from custom_layers import *
#from cbof import *
#from LearnToPayAttention import AttentionVGG
#batch_size=24
'''
class WildcatPool2d(Layer):
# initialize the layer, and set an extra parameter axis. No need to include inputs parameter
def __init__(self,kmax=0.2,kmin=0.2,alpha=0.7, **kwargs):
#self.axis = axis
self.kmax = kmax
self.kmin = kmin
self.alpha = alpha
self.result = None
super(WildcatPool2d, self).__init__(**kwargs)
# first use build function to define parameters, Creates the layer weights.
# input_shape will automatic collect input shapes to build layer
def build(self, input_shape):
#print(input_shape)
super(WildcatPool2d, self).build(input_shape)
def get_positive_k(self, k, n):
if k <= 0:
return 0
elif k < 1:
return K.cast(K.round(K.cast(n, dtype="float32")*
K.cast(k, dtype="float32")),dtype="int32")
elif k > n:
return n
else:
return int(k)
# This is where the layer's logic lives. In this example, I just concat two tensors.
def call(self, x, **kwargs):
batch_size, h, w, num_channels = K.shape(x)[0],K.shape(x)[1],K.shape(x)[2],K.shape(x)[3]
n = h * w # number of regions
kmax = self.get_positive_k(self.kmax, n)
kmin = self.get_positive_k(self.kmin, n)
x = K.reshape(x,(batch_size,n,num_channels))
x = K.permute_dimensions(x,(0,2,1))
x = tf.contrib.framework.sort(x,axis=-1,direction='DESCENDING')
x_max = K.sum(x[:,:,:kmax],axis=-1,keepdims=False)/K.cast(kmax,dtype="float32")
x_min = (K.sum(x[:,:,n-kmin:n],axis=-1,keepdims=False)
*self.alpha / K.cast(kmin,dtype="float32"))
self.result = Average()([x_max,x_min])
return self.result
# return output shape
def compute_output_shape(self, input_shape):
#return K.int_shape(self.result)#(batch_size,num_classes)
return tuple([input_shape[0],input_shape[3]])
'''
#################################################################
def target_category_loss(x, category_index, nb_classes):
#batch_label=K.zeros((K.shape(x)[0],nb_classes))
#batch_label=batch_label[:,category_index].assign(K.ones((K.shape(x)[0],)))
batch_label=K.zeros((batch_size,nb_classes))
batch_label=batch_label[:,category_index].assign(K.ones((batch_size,)))
return tf.multiply(x, batch_label)
def target_category_loss_output_shape(input_shape):
return input_shape
def normalize(x):
# utility function to normalize a tensor by its L2 norm
return x / (K.sqrt(K.mean(K.square(x),axis=(1,2,3),keepdims=True)) + 1e-5)
class Get_grads(Layer):
def __init__(self, **kwargs):
#self.axis = axis
self.result = None
super(Get_grads, self).__init__(**kwargs)
def build(self, input_shape):
print(input_shape)
super(Get_grads, self).build(input_shape)
def call(self, x, **kwargs):
self.result = normalize(K.gradients(x[0], x[1])[0])
return self.result
def compute_output_shape(self, input_shape):
return K.int_shape(self.result)
# 冻上base_model所有层,这样就可以正确获得bottleneck特征
def setup_to_transfer_learn(base_model):
"""Freeze all layers and compile the model"""
for layer in base_model.layers:
layer.trainable = False
'''
def lr_multiply(base_model):
for layer in base_model.layers:
layer.W_learning_rate_multiplier = args_dict.lrmult_conv
layer.b_learning_rate_multiplier = args_dict.lrmult_conv
'''
class ModelFactory:
"""
Model facotry for Keras default models
"""
def __init__(self):
self.models_ = dict(
VGG16=dict(
input_shape=(224, 224, 3),
module_name="vgg16",
last_conv_layer="block5_conv3",
),
VGG19=dict(
input_shape=(224, 224, 3),
module_name="vgg19",
last_conv_layer="block5_conv4",
),
DenseNet121=dict(
input_shape=(224, 224, 3),
module_name="densenet",
last_conv_layer="bn",
),
DenseNet169=dict(
input_shape=(224, 224, 3),
module_name="densenet",
last_conv_layer="bn",
),
ResNet50=dict(
input_shape=(224, 224, 3),
module_name="resnet50",
last_conv_layer="activation_49",
),
InceptionV3=dict(
input_shape=(299, 299, 3),
module_name="inception_v3",
last_conv_layer="mixed10",
),
InceptionResNetV2=dict(
input_shape=(299, 299, 3),
module_name="inception_resnet_v2",
last_conv_layer="conv_7b_ac",
),
NASNetMobile=dict(
input_shape=(224, 224, 3),
module_name="nasnet",
last_conv_layer="activation_188",
),
NASNetLarge=dict(
input_shape=(331, 331, 3),
module_name="nasnet",
last_conv_layer="activation_260",
),
DarkNet19_448=dict(
input_shape=(224, 224, 3),
module_name="darknet19_448",
last_conv_layer="activation_260",
),
Xception=dict(
input_shape=(299, 299, 3),
module_name="xception",
last_conv_layer="activation_260",
),
)
def get_last_conv_layer(self, model_name):
return self.models_[model_name]["last_conv_layer"]
def get_input_size(self, model_name):
return self.models_[model_name]["input_shape"][:2]
def get_model(self, class_names, model_name="DenseNet121"
, use_base_weights=True, weights_path=None
, input_shape=None, model_id=7):
if use_base_weights is True:
base_weights = "imagenet"
else:
base_weights = None
base_model_class = getattr(
importlib.import_module(
#f"keras.applications.{self.models_[model_name]['module_name']}"
"keras.applications."+self.models_[model_name]['module_name']
),
model_name)
if input_shape is None:
input_shape = self.models_[model_name]["input_shape"]
img_input = Input(shape=input_shape)
base_model = None
base_model = base_model_class(
include_top=False,
input_tensor=img_input,
input_shape=input_shape,
weights=base_weights,
pooling="avg")
'''
train bcnn with two steps:
1.freeze base models,only train bilinear pooling and last fc layers with high lr=0.01
2.train all layers with lr=0.001
'''
#setup_to_transfer_learn(base_model)
layer_dict = dict([(layer.name, layer) for layer in base_model.layers])
conv_outputs = None #last conv output
if model_name=="VGG16":
block4_conv3 = layer_dict["block4_conv3"]
block4_conv3_outputs = block4_conv3.output
final_conv_layer = layer_dict["block5_conv3"]
conv_outputs = final_conv_layer.output
if model_name=="DenseNet121" or model_name=="DenseNet169":
final_conv_layer = layer_dict["bn"]
conv_outputs = final_conv_layer.output
if model_name=="InceptionV3":
final_conv_layer = layer_dict["mixed10"]
conv_outputs = final_conv_layer.output
if model_id == 0:
x = base_model.output
'''x = conv_outputs
##############SE module####################
squeeze = GlobalAveragePooling2D()(x)
excitation = Dense(units=512 // 4, activation='relu')(squeeze)
#excitation = Activation('relu')(excitation)
excitation = Dense(units=512, activation='sigmoid')(excitation)
#excitation = Activation('sigmoid')(excitation)
excitation = Reshape((1,1,512))(excitation)
x = Multiply()([x,excitation])
#x = SqueezeExcitation(512)(x)
###########################################
spatial_att = Conv2D(128, (1, 1), activation='relu', padding='same', name='conv6')(x)
spatial_att = Conv2D(1, (1, 1), activation='sigmoid', padding='same', name='loc')(spatial_att)
x = Multiply()([x,spatial_att])
x = GlobalAveragePooling2D()(x)'''
predictions = Dense(len(class_names), activation="softmax", name="cls_pred")(x)
elif model_id == 1:
loc = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp0')(conv_outputs)
#conv6 = LocallyConnected2D(32, (3, 3), activation='relu', padding='valid', name='conv6')(cccp)
loc = Conv2D(128, (1, 1), activation='relu', padding='same', name='conv6')(loc)
loc = Conv2D(1, (1, 1), activation='relu', padding='same', name='loc')(loc)
x = base_model.output
#x = conv_outputs
#x = x * loc
#AttributeError: 'Tensor' object has no attribute '_keras_history'此处不能用后端函数
#x = Multiply()([x,loc])
#x = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp1')(x)
#x = Conv2D(128, (3, 3), activation='relu', padding='same', name='conv7')(x)
#x = GlobalAveragePooling2D()(x)
#x = GlobalMaxPooling2D()(x)
#x = Dropout(rate=0.5)(x)
predictions = Dense(len(class_names), activation="softmax", name="cls_pred")(x)
elif model_id == 2:
#x = base_model.output
x = conv_outputs
#x = Multiply()([x,loc])
z_l2=BilinearPooling()(x)
#x = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp1')(x)
#x = Conv2D(128, (3, 3), activation='relu', padding='same', name='conv7')(x)
#x = GlobalAveragePooling2D()(x)
#x = GlobalMaxPooling2D()(x)
predictions = Dense(len(class_names), activation="softmax", name="cls_pred")(z_l2)
#freeze_model = Model(inputs=img_input, output=predictions)
#setup_to_transfer_learn(freeze_model)
loc = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp0')(conv_outputs)
loc = Conv2D(128, (1, 1), activation='relu', padding='same', name='conv6')(loc)
loc = Conv2D(1, (1, 1), activation='relu', padding='same', name='loc')(loc)
elif model_id == 3:
loc0 = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp0')(conv_outputs)
loc0 = Conv2D(128, (1, 1), activation='relu', padding='same', name='conv6')(loc0)
loc0 = Conv2D(1, (1, 1), activation='relu', padding='same', name='loc0')(loc0)
loc1 = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp1')(block4_conv3_outputs)
loc1 = Conv2D(256, (1, 1), activation='relu', padding='same', name='conv7')(loc1)
loc1 = Conv2D(1, (1, 1), activation='relu', padding='same', name='loc1')(loc1)
my_resize1 = Lambda(lambda x: K.repeat_elements(x, 2, axis=1))
x = conv_outputs
x_att = Multiply()([x,loc0])
loc0 = my_resize1(loc0)
my_resize2 = Lambda(lambda x: K.repeat_elements(x, 2, axis=2))
loc0 = my_resize2(loc0)
#loc = Add(name='loc')([loc0, loc1])
loc = Average(name='loc')([loc0, loc1])
#x = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp1')(x)
#x = Conv2D(128, (3, 3), activation='relu', padding='same', name='conv7')(x)
#x = GlobalAveragePooling2D()(x)
x1 = GlobalMaxPooling2D()(x_att)
#x = Dropout(rate=0.5)(x)
predictions = Dense(len(class_names), activation="softmax", name="cls_pred")(x1)
x = my_resize1(x)
x = my_resize2(x)
x_merge = Concatenate(axis=-1)([x,block4_conv3_outputs])
x_att1 = Multiply()([x_merge,loc])
x2 = GlobalMaxPooling2D()(x_att1)
predictions1 = Dense(len(class_names), activation="softmax", name="cls_pred1")(x2)
elif model_id == 4:
loc = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp')(conv_outputs)
loc = Conv2D(256, (1, 1), activation='relu', padding='same', name='conv6')(loc)
loc = Conv2D(1, (1, 1), activation='relu', padding='same', name='loc')(loc)
#TypeError: Output tensors to a Model must be Keras tensors. Found: Tensor("Squeeze:0", shape=(?, 14, 14), dtype=float32)
#loc = K.squeeze(loc,axis=3)
x = conv_outputs
x = AveragePooling2D(pool_size=(2, 2))(x)
#x = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp2')(x)
#x = Conv2D(128, (3, 3), activation='relu', padding='same', name='conv7')(x)
#x = GlobalAveragePooling2D()(x)
#x = GlobalMaxPooling2D()(x)
x = NoisyAnd()(x)
#x = GlobalMaxPooling2D()(x)
#print predictions.shape
#my_reshape = Lambda(lambda x: K.reshape(x, (-1, x.shape[3])))
#x = my_reshape(x)
predictions = Dense(len(class_names), activation="softmax", name="cls_pred")(x)
elif model_id == 5:
loc = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp0')(conv_outputs)
loc = Conv2D(128, (1, 1), activation='relu', padding='same', name='conv6')(loc)
loc = Conv2D(1, (1, 1), activation='relu', padding='same', name='loc')(loc)
#x = base_model.output
x = conv_outputs
x = AveragePooling2D(pool_size=(2, 2))(x)
x1 = AveragePooling2D(pool_size=(2, 2))(x)
#x = Multiply()([x,loc])
#x = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp1')(x)
#x = Conv2D(128, (3, 3), activation='relu', padding='same', name='conv7')(x)
x = Conv2D(2, (1, 1), activation='relu', padding='same', name='cccp1')(x)
x1 = Conv2D(2, (1, 1), activation='relu', padding='same', name='cccp2')(x1)
#x = GlobalAveragePooling2D()(x)
#x = GlobalMaxPooling2D()(x)
x = Softmax4D()(x)
#x = GlobalMaxPooling2D()(x)
x1 = Softmax4D()(x1)
x = MaxPooling2D(pool_size=(14, 14))(x)
x1 = MaxPooling2D(pool_size=(7, 7))(x1)
x = Flatten(name='flatten')(x)
x1 = Flatten(name='flatten1')(x1)
predictions = Recalc(axis=1, name='cls_pred0')(x)
predictions1 = Recalc(axis=1, name='cls_pred1')(x1)
#predictions1 = Recalc(axis=1)(x1)
predictions = Average(name='cls_pred')([predictions, predictions1])
#predictions = Dense(len(class_names), activation="softmax", name="cls_pred")(x)
elif model_id == 6:
x = base_model.output
pred = Dense(len(class_names), activation="softmax", name="pred")(x)
target_layer = Lambda(lambda x: target_category_loss(x, 1, 2),output_shape = target_category_loss_output_shape)
gc = target_layer(pred)
get_loss = Lambda(lambda x: K.sum(x,axis=1))
loss = get_loss(gc)
grads = Get_grads()([loss, conv_outputs])
get_weights = Lambda(lambda x: K.mean(x, axis = (1, 2),keepdims=True))
weights = get_weights(grads)
my_resize1 = Lambda(lambda x: K.repeat_elements(x, conv_outputs.shape[1], axis=1))
weights = my_resize1(weights)
my_resize2 = Lambda(lambda x: K.repeat_elements(x, conv_outputs.shape[2], axis=2))
weights = my_resize2(weights)
grad_cam = Multiply()([conv_outputs,weights])
loc = Conv2D(128, (1, 1), activation='relu', padding='same', name='conv6')(conv_outputs)
loc = Conv2D(1, (1, 1), activation='relu', padding='same', name='loc')(loc)
#x = base_model.output
#x = Multiply()([x,loc])
#x = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp1')(x)
#x = Conv2D(128, (3, 3), activation='relu', padding='same', name='conv7')(x)
#x = GlobalAveragePooling2D()(x)
#x = GlobalMaxPooling2D()(x)
x = GlobalMaxPooling2D()(grad_cam)
#x1 = MaxPooling2D(pool_size=(7, 7))(x1)
#x = Flatten(name='flatten')(x)
predictions = Dense(len(class_names), activation="softmax", name="cls_pred")(x)
elif model_id == 7:
#loc = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp0')(conv_outputs)
#loc = Conv2D(128, (1, 1), activation='relu', padding='same', name='conv6')(loc)
#loc = Conv2D(1, (1, 1), activation='relu', padding='same', name='loc')(loc)
x = conv_outputs
#x = Multiply()([x,loc])
#num_maps=8
classes=2
#x = Conv2D(num_maps*classes, (1, 1), activation='relu', padding='same', name='cccp')(x)
#x = ClassWisePool()(x)
x = WildcatPool2d()(x)
#x = LogSumExp()(x)
#predictions = Recalc(axis=1, name='cls_pred')(x)#
predictions = Dense(len(class_names), activation="softmax", name="cls_pred")(x)
#predictions = Dense(len(class_names), activation='sigmoid', name='cls_pred')(x)
elif model_id == 8:
#loc = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp0')(conv_outputs)
#loc = Conv2D(128, (1, 1), activation='relu', padding='same', name='conv6')(loc)
#loc = Conv2D(1, (1, 1), activation='relu', padding='same', name='loc')(loc)
x = conv_outputs
#x = Multiply()([x,loc])
x = LogSumExp(r=1)(x)
predictions = Dense(len(class_names), activation="softmax", name="cls_pred")(x)
elif model_id == 9:
loc = Conv2D(512, (1, 1), activation='relu', padding='same', name='cccp0')(conv_outputs)
loc = Conv2D(128, (1, 1), activation='relu', padding='same', name='conv6')(loc)
loc = Conv2D(1, (1, 1), activation='relu', padding='same', name='loc')(loc)
x = conv_outputs
n_codewords=128
x=BoF_Pooling(n_codewords, spatial_level=0)(x)
predictions = Dense(len(class_names), activation="softmax", name="cls_pred")(x)
elif model_id == 10:
base_model=AttentionVGG(img_input, outputclasses=2, batchnorm=False, batchnormalizeinput=False).model
model = Model(inputs=img_input, output=#base_model.output#predictions,#predictions1,
predictions
#loc
)
if weights_path == "":
weights_path = None
if weights_path is not None:
#print(f"load model weights_path: {weights_path}")
print ("load model weights_path: {}".format(weights_path))
model.load_weights(weights_path, by_name=True)
return model