def make_model(num_classes, name='midi', **kwargs):
input_shape = (25, 25, 25)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
model = Sequential()
#model = all_preprocessing(model, 'all', input_shape=k_input_shape, **kwargs)
model = all_preprocessing(model, ['normalize'],
input_shape=k_input_shape,
**kwargs)
model.add(Flatten())
model.add(Dense(150))
model.add(Activation('relu'))
model.add(Dropout(0.75))
model.add(Dense(150))
model.add(Activation('relu'))
model.add(Dropout(0.75))
model.add(Dense(num_classes))
model.add(Activation('softmax'))
return model, name, input_shape
def make_model(num_classes, **model_params):
input_shape = (13, 13, 13)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
if 'normalize' in model_params:
normalize = model_params['normalize']
else:
normalize = True
if 'dense_dropout_p' in model_params:
dense_dropout_p = model_params['dense_dropout_p']
else:
dense_dropout_p = 0.25
model = Sequential()
model = all_preprocessing(model,
normalize=True,
input_shape=k_input_shape,
**model_params)
model.add(Flatten())
model.add(Dense(150))
model.add(Activation('relu'))
model.add(Dropout(dense_dropout_p))
model.add(Dense(num_classes))
model.add(Activation('softmax'))
return model, input_shape
def make_model(num_classes, name='conv_2_layer', **model_params):
if not 'normalize' in model_params:
model_params['normalize'] = True
if not 'rotation' in model_params:
model_params['rotation'] = False
if not 'foveation' in model_params:
model_params['foveation'] = False
if not 'noise' in model_params:
model_params['noise'] = False
if not 'linear_deformation' in model_params:
model_params['linear_deformation'] = False
if not 'non_linear_resampling' in model_params:
model_params['non_linear_resampling'] = False
if 'conv_dropout_p' in model_params:
conv_dropout_p = model_params['conv_dropout_p']
else:
conv_dropout_p = 0.0
if 'dense_dropout_p' in model_params:
dense_dropout_p = model_params['dense_dropout_p']
else:
dense_dropout_p = 0.0
name = name + '_' + str(conv_dropout_p) + '_' + str(dense_dropout_p)
input_shape = (25, 25, 25)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
model = Sequential()
model = all_preprocessing(model, input_shape=k_input_shape, **model_params)
model.add(Conv3D(48, (5, 5, 5), padding='valid'))
model.add(Activation('relu'))
model.add(MaxPooling3D(pool_size=(2, 2, 2)))
model.add(Dropout(conv_dropout_p))
model.add(Conv3D(64, (4, 4, 4), padding='valid'))
model.add(Activation('relu'))
model.add(MaxPooling3D(pool_size=(2, 2, 2)))
model.add(Dropout(conv_dropout_p))
model.add(Flatten())
model.add(Dense(150))
model.add(Activation('relu'))
model.add(Dropout(dense_dropout_p))
model.add(Dense(num_classes))
model.add(Activation('softmax')) #softmax
return model, input_shape
def make_model(num_classes,
name='test_foveation',
**kwargs):
input_shape = (25, 25, 25)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
model = Sequential()
model = all_preprocessing(model, foveation=True, input_shape=k_input_shape, **kwargs)
return model, name, input_shape
def make_model(num_classes, **kwargs):
input_shape = (25, 25, 25)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
model = Sequential()
model = all_preprocessing(model,
normalize=True,
input_shape=k_input_shape,
**kwargs)
return model, input_shape
def make_model(num_classes, name='test_rotate', **kwargs):
input_shape = (25, 25, 25)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
model = Sequential()
model = all_preprocessing(model,
rotation=True,
input_shape=k_input_shape,
force_use_in_test_phase=True,
**kwargs)
return model, name, input_shape
def make_model(num_classes,
conv_dropout_p=0.75,
dense_dropout_p=0.5,
name='conv_2_layer_pass_through',
**kwargs):
name = name + '_' + str(conv_dropout_p) + '_' + str(dense_dropout_p)
input_shape = (25, 25, 25)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
inputs = Input(shape=(input_shape[0], input_shape[1], input_shape[2], 1))
processed = all_preprocessing(inputs, 'all', functional_api=True, **kwargs)
conv_1 = Conv3D(48, (5, 5, 5), padding='valid')(processed)
conv_1 = Activation('relu')(conv_1)
pool_1 = MaxPooling3D(pool_size=(3, 3, 3))(conv_1)
drop_1a = Dropout(conv_dropout_p)(pool_1)
crop_1 = Cropping3D((7, 7, 7))(conv_1)
drop_1b = Dropout(conv_dropout_p)(crop_1)
conc_1 = Concatenate(axis=4)([drop_1a, drop_1b])
conv_2 = Conv3D(96, (3, 3, 3), padding='valid')(conc_1)
conv_2 = Activation('relu')(conv_2)
pool_2 = MaxPooling3D(pool_size=(3, 3, 3))(conv_2)
drop_2a = Dropout(conv_dropout_p)(pool_2)
crop_2 = Cropping3D((2, 2, 2))(conv_2)
drop_2b = Dropout(conv_dropout_p)(crop_2)
conc_2 = Concatenate(axis=4)([drop_2a, drop_2b])
flat = Flatten()(conc_2)
fc_1 = Dense(150)(flat)
fc_1 = Activation('relu')(fc_1)
drop_3 = Dropout(dense_dropout_p)(fc_1)
fc_2 = Dense(num_classes)(drop_3)
predictions = Activation('softmax')(fc_2)
model = Model(inputs=inputs, outputs=predictions)
return model, name, input_shape
def make_model(num_classes, name='test_non_linear_resampling', **kwargs):
input_shape = (25, 25, 25)
target_shape = (15, 15, 15)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
model = Sequential()
model = all_preprocessing(model,
non_linear_resampling=True,
input_shape=k_input_shape,
target_shape=target_shape,
force_use_in_test_phase=True,
**kwargs)
model.add(Cropping3D((5, 5, 5)))
return model, name, input_shape
def make_model(num_classes, name='conv_2_layer_non_linear', **kwargs):
conv_dropout_p = 0.5
dense_dropout_p = 0.5
input_shape = (25, 25, 25)
target_shape = (15, 15, 15)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
model = Sequential()
model = all_preprocessing(model,
normalize=True,
rotation=True,
foveation=False,
noise=True,
linear_deformation=False,
non_linear_resampling=True,
input_shape=k_input_shape,
target_shape=target_shape,
force_use_in_test_phase=False,
**kwargs)
model.add(Cropping3D((5, 5, 5)))
model.add(Conv3D(16, (7, 7, 7), padding='valid'))
model.add(Activation('relu'))
model.add(Dropout(conv_dropout_p))
model.add(Conv3D(16, (7, 7, 7), padding='valid'))
model.add(Activation('relu'))
model.add(Dropout(conv_dropout_p))
model.add(Flatten())
model.add(Dense(50))
model.add(Activation('relu'))
model.add(Dropout(dense_dropout_p))
model.add(Dense(num_classes))
model.add(Activation('softmax')) #softmax
return model, name, input_shape
def make_model(num_classes, **kwargs):
input_shape = (25, 25, 25)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
model = Sequential()
model = all_preprocessing(
model,
normalize=True,
rotation=True,
noise=True,
input_shape=k_input_shape,
**kwargs,
)
model.add(Flatten())
model.add(Dense(num_classes))
model.add(Activation('softmax'))
return model, input_shape
def make_model(num_classes,
conv_dropout_p=0.75,
dense_dropout_p=0.5,
name='conv_2_layer_no_pool',
**kwargs):
name = name + '_' + str(conv_dropout_p) + '_' + str(dense_dropout_p)
input_shape = (25, 25, 25)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
model = Sequential()
model = all_preprocessing(model,
'all',
input_shape=k_input_shape,
**kwargs)
model.add(Conv3D(48, (5, 5, 5), padding='valid'))
model.add(Activation('relu'))
model.add(Cropping3D(((5, 5), (5, 5), (5, 5))))
model.add(Dropout(conv_dropout_p))
model.add(Conv3D(48, (3, 3, 3), padding='valid'))
model.add(Activation('relu'))
model.add(Cropping3D(((3, 3), (3, 3), (3, 3))))
model.add(Dropout(conv_dropout_p))
model.add(Flatten())
model.add(Dense(150))
model.add(Activation('relu'))
model.add(Dropout(dense_dropout_p))
model.add(Dense(num_classes))
model.add(Activation('softmax'))
return model, name, input_shape
def make_model(num_classes,
rotation,
foveation,
noise,
linear_deformation,
non_linear_resampling,
name='conv_2_layer_conf',
**kwargs):
conv_dropout_p = 0.5
dense_dropout_p = 0.5
if non_linear_resampling:
input_shape = (25, 25, 25)
else:
input_shape = (15, 15, 15)
target_shape = (15, 15, 15)
k_input_shape = (input_shape[0], input_shape[1], input_shape[2], 1)
model = Sequential()
if rotation:
name += '_rotation'
if foveation:
name += '_foveation'
if noise:
name += '_noise'
if linear_deformation:
name += '_linear_deformation'
if non_linear_resampling:
name += '_non_linear_resampling'
model = all_preprocessing(model,
normalize=True,
rotation=rotation,
foveation=foveation,
noise=noise,
linear_deformation=linear_deformation,
non_linear_resampling=non_linear_resampling,
input_shape=k_input_shape,
target_shape=target_shape,
**kwargs)
if non_linear_resampling:
model.add(Cropping3D((5, 5, 5)))
model.add(Conv3D(32, (7, 7, 7), padding='valid'))
model.add(Activation('relu'))
model.add(Dropout(conv_dropout_p))
model.add(Conv3D(32, (7, 7, 7), padding='valid'))
model.add(Activation('relu'))
model.add(Dropout(conv_dropout_p))
model.add(Flatten())
model.add(Dense(100))
model.add(Activation('relu'))
model.add(Dropout(dense_dropout_p))
model.add(Dense(num_classes))
model.add(Activation('softmax')) #softmax
return model, name, input_shape
