def resume_routine_vgg(config,
weights_path,
verbose=(2, 2),
nb_epoch_finetune=15,
nb_epoch_after=50,
stiefel_observed=None,
stiefel_lr=0.01,
by_name=False):
"""
Finetune the ResNet-DCov
Returns
-------
"""
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
)
resume_model_with_config(VGG16_o2,
config,
weights_path=weights_path,
title='minc2500_VGG',
verbose=verbose,
image_gen=image_gen,
nb_epoch_after=nb_epoch_after,
nb_epoch_finetune=nb_epoch_finetune,
stiefel_lr=stiefel_lr,
stiefel_observed=stiefel_observed,
by_name=by_name)
def test_routine_resnet(config):
"""
Finetune the ResNet-DCov
Returns
-------
"""
nb_epoch_finetune = 2
nb_epoch_after = 50
title = 'test_minc_orig_cov'
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
# preprocessing_function=preprocess_image_for_imagenet
# channelwise_std_normalization=True
)
run_finetune(ResNet50_o2,
mincorig_finetune,
input_shape,
config,
image_gen,
nb_classes=nb_classes,
nb_epoch_finetune=nb_epoch_finetune,
nb_epoch_after=nb_epoch_after,
title=title,
verbose=(2, 1))
def run_routine_resnet_multibranch(config,
verbose=(2, 2),
nb_epoch_finetune=15,
nb_epoch_after=50,
stiefel_observed=None,
stiefel_lr=0.01):
"""
Finetune the ResNet-DCov
Returns
-------
"""
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
preprocessing_function=preprocess_image_for_imagenet)
run_model_with_config(ResNet50_o2_multibranch,
config,
title='minc_orig_resnet_MB',
verbose=verbose,
image_gen=image_gen,
nb_epoch_after=nb_epoch_after,
nb_epoch_finetune=nb_epoch_finetune,
stiefel_lr=stiefel_lr,
stiefel_observed=stiefel_observed)
def baseline_finetune_resnet(exp):
nb_epoch_finetune = 10
nb_epoch_after = 50
if exp == 1:
model = ResNet50_o1(denses=[],
nb_classes=nb_classes,
input_shape=input_shape,
freeze_conv=True)
model.name = 'baseline_resnet'
elif exp == 2:
last_conv_feature_maps = [1024, 512]
model = ResNet50_o1(denses=[],
nb_classes=nb_classes,
input_shape=input_shape,
freeze_conv=True,
last_conv_feature_maps=last_conv_feature_maps)
model.name = 'baseline_resnet_with_1x1'
else:
return
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
)
mincorig_finetune(model,
image_gen=image_gen,
title='mincorig_finetune_resnet',
nb_epoch_finetune=nb_epoch_finetune,
nb_epoch_after=nb_epoch_after,
batch_size=32,
early_stop=True)
def run_routine_resnet_multibranch(config,
verbose=(2, 2),
nb_epoch_finetune=15,
nb_epoch_after=50,
stiefel_observed=None,
stiefel_lr=0.01):
"""
Finetune the ResNet-DCov
Returns
-------
"""
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
)
finetune_model_with_config(ResNet50_o2_multibranch,
imagenet_finetune,
config,
nb_classes,
title='ImageNet_ResNet_MB',
verbose=verbose,
image_gen=image_gen,
nb_epoch_after=nb_epoch_after,
nb_epoch_finetune=nb_epoch_finetune,
stiefel_lr=stiefel_lr,
stiefel_observed=stiefel_observed)
def loadImageNet(
image_gen=None,
batch_size=32,
):
# Absolute paths
IMAGENET_PATH = '/home/kyu/.keras/datasets/ILSVRC2015'
TARGET_SIZE = (224, 224)
RESCALE_SMALL = 256
# ImageNet generator
imageNetLoader = ImageNetLoader(IMAGENET_PATH)
if image_gen is None:
iamge_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
preprocessing_function=preprocess_image_for_imagenet
# channelwise_std_normalization=True
)
train = imageNetLoader.generator('train',
image_data_generator=image_gen,
batch_size=batch_size)
valid = imageNetLoader.generator('valid',
image_data_generator=image_gen,
batch_size=batch_size)
# test = imageNetLoader.generator('valid', image_data_generator=gen)
return train, valid
def baseline_finetune_bilinear(exp=1):
nb_epoch_finetune = 1
nb_epoch_after = 100
if exp == 1:
model = VGG16_bilinear(nb_class=nb_classes,
input_shape=input_shape,
freeze_conv=True)
model.name = 'baseline_vgg16_bilinear'
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
preprocessing_function=preprocess_image_for_imagenet
# channelwise_std_normalization=True
)
weight_path = '/home/kyu/.keras/models/minc_finetune_vgg16_bilinear-baseline_vgg16_bilinear_1.weights_100'
minc2500_finetune(model,
image_gen=image_gen,
title='minc_finetune_vgg16_bilinear',
nb_epoch_finetune=nb_epoch_finetune,
nb_epoch_after=nb_epoch_after,
weight_path=weight_path,
load=True,
batch_size=32,
early_stop=True,
lr_decay=False,
lr=0.05)
def resume_routine_resnet_multibranch(config,
verbose=(2, 2),
weights_path='',
nb_epoch_finetune=0,
nb_epoch_after=50,
stiefel_observed=None,
stiefel_lr=0.01,
by_name=False):
"""
Finetune the ResNet-DCov
Returns
-------
"""
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
)
resume_model_with_config(ResNet50_o2_multibranch,
config,
weights_path=weights_path,
title='mit_indoor_resnet_MB',
verbose=verbose,
image_gen=image_gen,
nb_epoch_after=nb_epoch_after,
nb_epoch_finetune=nb_epoch_finetune,
stiefel_lr=stiefel_lr,
stiefel_observed=stiefel_observed,
by_name=by_name)
def run_routine_resnet(config, verbose=(2,2), nb_epoch_finetune=15, nb_epoch_after=50,
stiefel_observed=None, stiefel_lr=0.01):
"""
Finetune the ResNet-DCov
Returns
-------
"""
image_gen = ImageDataGeneratorAdvanced(TARGET_SIZE, RESCALE_SMALL, True,
horizontal_flip=True,
)
monitor_class = (O2Transform, SecondaryStatistic)
# monitor_metrics = ['weight_norm',]
# monitor_metrics = ['output_norm',]
monitor_metrics = ['matrix_image',]
if stiefel_observed is None:
run_finetune(ResNet50_o2, mincorig_finetune,
nb_classes=nb_classes,
input_shape=input_shape, config=config,
nb_epoch_finetune=nb_epoch_finetune, nb_epoch_after=nb_epoch_after,
image_gen=image_gen, title='minc_orig_resnet50', verbose=verbose,
monitor_classes=monitor_class,
monitor_measures=monitor_metrics)
else:
run_finetune_with_Stiefel_layer(ResNet50_o2, mincorig_finetune,
nb_classes=nb_classes,
input_shape=input_shape, config=config,
nb_epoch_finetune=nb_epoch_finetune, nb_epoch_after=nb_epoch_after,
image_gen=image_gen, title='minc_orig_resnet50_stiefel', verbose=verbose,
monitor_classes=monitor_class,
monitor_measures=monitor_metrics,
observed_keywords=stiefel_observed,
lr=stiefel_lr)
def baseline_finetune_vgg(exp=1):
nb_epoch_finetune = 10
nb_epoch_after = 50
if exp == 1:
model = VGG16_o1(denses=[4096, 4096],
nb_classes=nb_classes,
input_shape=input_shape,
freeze_conv=True,
last_conv=False)
model.name = 'baseline_vgg16'
elif exp == 2:
model = VGG16_o1(denses=[4096, 4096],
nb_classes=nb_classes,
input_shape=input_shape,
freeze_conv=True,
last_conv=True)
model.name = 'baseline_vgg16_with_1x1'
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
preprocessing_function=preprocess_image_for_imagenet
# channelwise_std_normalization=True
)
minc2500_finetune(model,
image_gen=image_gen,
title='minc2500_finetune_vgg16',
nb_epoch_finetune=nb_epoch_finetune,
nb_epoch_after=nb_epoch_after,
batch_size=32,
early_stop=True)
def run_residual_cov_resnet(exp):
"""
Finetune the ResNet-DCov
Returns
-------
m
"""
nb_epoch_finetune = 50
nb_epoch_after = 50
config = get_residual_cov_experiment(exp)
title = 'minc2500_residual_cov'
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
# preprocessing_function=preprocess_image_for_imagenet
# channelwise_std_normalization=True
)
run_finetune(ResNet50_o2,
minc2500_finetune,
input_shape,
config,
image_gen,
nb_classes=nb_classes,
nb_epoch_finetune=nb_epoch_finetune,
nb_epoch_after=nb_epoch_after,
title=title,
verbose=(2, 1))
def run_routine_vgg(config,
verbose=(2, 2),
nb_epoch_finetune=15,
nb_epoch_after=50,
stiefel_observed=None,
stiefel_lr=0.01):
"""
Finetune the ResNet-DCov
Returns
-------
"""
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
preprocessing_function=preprocess_image_for_imagenet)
monitor_class = (O2Transform, SecondaryStatistic)
# monitor_metrics = ['weight_norm',]
# monitor_metrics = ['output_norm',]
monitor_metrics = [
'matrix_image',
]
run_model_with_config(VGG16_o2,
config,
title='minc2500_VGG16',
verbose=verbose,
image_gen=image_gen,
nb_epoch_after=nb_epoch_after,
nb_epoch_finetune=nb_epoch_finetune,
stiefel_lr=stiefel_lr,
stiefel_observed=stiefel_observed)
def test_routine_vgg(exp=1):
"""
Finetune the ResNet-DCov
Returns
-------
"""
nb_epoch_finetune = 10
nb_epoch_after = 10
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
preprocessing_function=preprocess_image_for_imagenet
# channelwise_std_normalization=True
)
config = get_experiment_settings(exp)
cov_branch = config.cov_branch
cov_mode = config.cov_mode
cov_regularizer = config.cov_regularizer
early_stop = config.early_stop
print("Running experiment {}".format(exp))
for param in config.params:
for mode in config.mode_list:
for cov_output in config.cov_outputs:
print("Run routine 1 param {}, mode {}, covariance output {}".
format(param, mode, cov_output))
model = VGG16_o2(
parametrics=param,
mode=mode,
cov_branch=cov_branch,
cov_mode=cov_mode,
nb_classes=nb_classes,
cov_branch_output=cov_output,
input_shape=input_shape,
freeze_conv=True,
cov_regularizer=cov_regularizer,
last_conv_feature_maps=config.last_conv_feature_maps)
mincorig_finetune(model,
title='minc_orig_cov_{}_wv{}_{}'.format(
cov_branch, str(cov_output), cov_mode),
nb_epoch_after=nb_epoch_after,
nb_epoch_finetune=nb_epoch_finetune,
image_gen=image_gen,
batch_size=32,
early_stop=early_stop,
log=False,
verbose=1)
def baseline_VGG():
model = VGG16_o1(denses=[1024],
nb_classes=nb_classes,
input_shape=input_shape)
model.name = 'baseline_vgg16'
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
preprocessing_function=preprocess_image_for_imagenet
# channelwise_std_normalization=True
)
dtd_fitting(model, image_gen)
def run_model_with_config(model,
config,
title='minc2500',
image_gen=None,
verbose=(2, 2),
nb_epoch_finetune=15,
nb_epoch_after=50,
stiefel_observed=None,
stiefel_lr=0.01):
"""
Finetune the ResNet-DCov
Returns
-------
"""
if image_gen is None:
image_gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
)
monitor_class = (O2Transform, SecondaryStatistic)
# monitor_metrics = ['weight_norm',]
# monitor_metrics = ['output_norm',]
monitor_metrics = [
'matrix_image',
]
finetune_model_with_config(
model,
minc2500_finetune,
config,
nb_classes,
input_shape,
title=title,
image_gen=image_gen,
verbose=verbose,
nb_epoch_finetune=nb_epoch_finetune,
nb_epoch_after=nb_epoch_after,
stiefel_observed=stiefel_observed,
stiefel_lr=stiefel_lr,
)
def resume_model_with_config(model, config, title='cifar10',
image_gen=None, weights_path='',
verbose=(2, 2), nb_epoch_finetune=0, nb_epoch_after=50,
stiefel_observed=None, stiefel_lr=0.0001,
by_name=False):
"""
Finetune the ResNet-DCov
Returns
-------
"""
if image_gen is None:
image_gen = ImageDataGeneratorAdvanced(TARGET_SIZE, RESCALE_SMALL, True,
horizontal_flip=True,
)
monitor_class = (O2Transform, SecondaryStatistic)
# monitor_metrics = ['weight_norm',]
# monitor_metrics = ['output_norm',]
monitor_metrics = ['matrix_image', ]
if stiefel_observed is None:
resume_finetune(model, cifar_train,
weights_path=weights_path,
nb_classes=nb_classes,
input_shape=input_shape, config=config,
nb_epoch_finetune=nb_epoch_finetune, nb_epoch_after=nb_epoch_after,
image_gen=image_gen, title=title, verbose=verbose,
monitor_classes=monitor_class,
monitor_measures=monitor_metrics,
by_name=by_name)
else:
resume_finetune_with_Stiefel_layer(model, cifar_train,
weights_path=weights_path,
nb_classes=nb_classes,
input_shape=input_shape, config=config,
nb_epoch_finetune=nb_epoch_finetune, nb_epoch_after=nb_epoch_after,
image_gen=image_gen, title=title+'-stiefel', verbose=verbose,
monitor_classes=monitor_class,
monitor_measures=monitor_metrics,
observed_keywords=stiefel_observed,
lr=stiefel_lr,
by_name=by_name)
TARGET_SIZE = (224, 224)
RESCALE_SMALL = 256
BATCH_SIZE = 32
NB_EPOCH = 1000
VERBOSE = 1
SAVE_LOG = True
VALIDATION = True
# Xray Loader
xray_loader = XrayLoader()
gen = ImageDataGeneratorAdvanced(
TARGET_SIZE,
RESCALE_SMALL,
True,
horizontal_flip=True,
preprocessing_function=preprocess_image_for_imagenet
# channelwise_std_normalization=True
)
train = xray_loader.generator('train', batch_size=BATCH_SIZE)
test = xray_loader.generator('test', batch_size=BATCH_SIZE)
# train = xray_loader.generator('train', batch_size=BATCH_SIZE, image_data_generator=gen)
# test = xray_loader.generator('test', batch_size=BATCH_SIZE, image_data_generator=gen)
def fit_model(model, load=False, save=True, title='swisscom', nb_epoch=10):
# sgd = SGD(lr=0.0001, decay=1e-6, momentum=0.9, nesterov=False)
sgd = SGDWithWeightnorm(lr=0.0001, decay=1e-6, momentum=0.9)
model.compile(loss='categorical_crossentropy',
