def __init__(self,
num_classes,
w_initializer=None,
w_regularizer=None,
b_initializer=None,
b_regularizer=None,
acti_func='leaky-relu',
name='UNet2D'):
BaseNet.__init__(self, num_classes=num_classes, name=name)
self.n_fea = [64, 128, 256, 512, 1024]
net_params = {
'padding': 'SAME',
'with_bias': True,
'with_bn': False,
'acti_func': acti_func,
'w_initializer': w_initializer,
'b_initializer': b_initializer,
'w_regularizer': w_regularizer,
'b_regularizer': b_regularizer
}
self.conv_params = {'kernel_size': 3, 'stride': 1}
self.deconv_params = {'kernel_size': 2, 'stride': 2}
self.pooling_params = {'kernel_size': 2, 'stride': 2}
self.conv_params.update(net_params)
self.deconv_params.update(net_params)
def __init__(self,
decay,
affine_w_initializer=None,
affine_b_initializer=None,
disp_w_initializer=None,
disp_b_initializer=None,
acti_func='relu',
interp='linear',
boundary='replicate',
name='inet-hybrid-pre-warp'):
"""
:param decay: float, regularisation decay
:param affine_w_initializer: weight initialisation for affine registration network
:param affine_b_initializer: bias initialisation for affine registration network
:param disp_w_initializer: weight initialisation for dense registration network
:param disp_b_initializer: bias initialisation for dense registration network
:param acti_func: activation function to use
:param interp: string, type of interpolation for the resampling [default:linear]
:param boundary: string, padding mode to deal with image boundary
:param name: layer name
"""
BaseNet.__init__(self, name=name)
self.global_net = INetAffine(decay=decay,
affine_w_initializer=affine_w_initializer,
affine_b_initializer=affine_b_initializer,
acti_func=acti_func,
name='inet-global')
self.local_net = INetDense(decay=decay,
disp_w_initializer=disp_w_initializer,
disp_b_initializer=disp_b_initializer,
acti_func=acti_func,
name='inet-local')
self.interp = interp
self.boundary = boundary
def __init__(self,
decay=1e-6,
affine_w_initializer=None,
affine_b_initializer=None,
acti_func='relu',
name='inet-affine'):
"""
This network estimates affine transformations from
a pair of moving and fixed image:
Hu et al., Label-driven weakly-supervised learning for
multimodal deformable image registration, arXiv:1711.01666
https://arxiv.org/abs/1711.01666
:param decay:
:param affine_w_initializer:
:param affine_b_initializer:
:param acti_func:
:param name:
"""
BaseNet.__init__(self, name=name)
self.fea = [4, 8, 16, 32, 64]
self.k_conv = 3
self.affine_w_initializer = affine_w_initializer
self.affine_b_initializer = affine_b_initializer
self.res_param = {
'w_initializer': GlorotUniform.get_instance(''),
'w_regularizer': regularizers.l2_regularizer(decay),
'acti_func': acti_func}
self.affine_param = {
'w_regularizer': regularizers.l2_regularizer(decay),
'b_regularizer': None}
def __init__(self,
num_classes,
w_initializer=None,
w_regularizer=None,
b_initializer=None,
b_regularizer=None,
acti_func='relu',
name='UNet2D'):
BaseNet.__init__(self,
num_classes=num_classes,
name=name)
self.n_fea = [64, 128, 256, 512, 1024]
net_params = {'padding': 'VALID',
'with_bias': True,
'with_bn': False,
'acti_func': acti_func,
'w_initializer': w_initializer,
'b_initializer': b_initializer,
'w_regularizer': w_regularizer,
'b_regularizer': b_regularizer}
self.conv_params = {'kernel_size': 3, 'stride': 1}
self.deconv_params = {'kernel_size': 2, 'stride': 2}
self.pooling_params = {'kernel_size': 2, 'stride': 2}
self.conv_params.update(net_params)
self.deconv_params.update(net_params)
def __init__(self,
num_classes,
w_initializer=None,
w_regularizer=None,
b_initializer=None,
b_regularizer=None,
acti_func='relu',
name='UNet2D'):
BaseNet.__init__(self,
num_classes=num_classes,
name=name)
self.n_fea = [64, 128, 256, 512, 1024]
net_params = {'padding': 'VALID',
'with_bias': True,
'feature_normalization': 'batch',
'group_size': -1,
'acti_func': acti_func,
'w_initializer': w_initializer,
'b_initializer': b_initializer,
'w_regularizer': w_regularizer,
'b_regularizer': b_regularizer}
self.conv_params = {'kernel_size': 3, 'stride': 1}
self.deconv_params = {'kernel_size': 2, 'stride': 2}
self.pooling_params = {'kernel_size': 2, 'stride': 2}
self.conv_params.update(net_params)
self.deconv_params.update(net_params)
def __init__(self,
decay=1e-6,
affine_w_initializer=None,
affine_b_initializer=None,
acti_func='relu',
name='inet-affine'):
"""
:param decay: float, regularisation decay
:param affine_w_initializer: weight initialisation for affine registration network
:param affine_b_initializer: bias initialisation for affine registration network
:param acti_func: activation function to use
:param name: layer name
"""
BaseNet.__init__(self, name=name)
self.fea = [4, 8, 16, 32, 64]
self.k_conv = 3
self.affine_w_initializer = affine_w_initializer
self.affine_b_initializer = affine_b_initializer
self.res_param = {
'w_initializer': GlorotUniform.get_instance(''),
'w_regularizer': regularizers.l2_regularizer(decay),
'acti_func': acti_func
}
self.affine_param = {
'w_regularizer': regularizers.l2_regularizer(decay),
'b_regularizer': None
}
def __init__(self,
decay=0.0,
smoothing=0,
disp_w_initializer=None,
disp_b_initializer=None,
acti_func='relu',
multi_scale_fusion=True,
name='inet-dense'):
"""
:param decay: float, regularisation decay
:param smoothing: float, smoothing factor for dense displacement field
:param disp_w_initializer: initialisation of the displacement fields
:param disp_b_initializer: initialisation of the displacement fields
:param acti_func: activation function to use
:param multi_scale_fusion: True/False indicating whether to use
multiscale feature fusion.
:param name: layer name
"""
BaseNet.__init__(self, name=name)
# self.fea = [40, 80, 160, 320, 640]
# self.fea = [32, 64, 128, 256, 512]
self.fea = [30, 60, 120, 240, 480]
# self.fea = [16, 32, 64, 128, 256]
self.k_conv = 3
self.multi_scale_fusion = multi_scale_fusion
self.down_res_param = {
'w_initializer': GlorotUniform.get_instance(''),
'w_regularizer': regularizers.l2_regularizer(decay),
'acti_func': acti_func}
self.up_res_param = {
'acti_func': acti_func,
'w_initializer': GlorotUniform.get_instance(''),
'w_regularizer': regularizers.l2_regularizer(decay),
'is_residual_upsampling': True,
'type_string': 'bn_acti_conv'}
# displacement initialiser & regulariser
if disp_w_initializer is None:
disp_b_initializer = tf.constant_initializer(0.0)
#disp_w_initializer = tf.random_normal_initializer(0, 1e-4)
if disp_b_initializer is None:
disp_b_initializer = tf.constant_initializer(0.0)
#disp_w_initializer = tf.random_normal_initializer(0, 0.0)
self.disp_param = {
'w_initializer': disp_w_initializer,
'w_regularizer': regularizers.l2_regularizer(decay),
'b_initializer': disp_b_initializer,
'b_regularizer': None}
if smoothing > 0:
self.smoothing_func = _smoothing_func(smoothing)
else:
self.smoothing_func = None
def __init__(self,
decay,
affine_w_initializer=None,
affine_b_initializer=None,
disp_w_initializer=None,
disp_b_initializer=None,
acti_func='relu',
interp='linear',
boundary='replicate',
name='inet-hybrid-pre-warp'):
"""
Re-implementation of the registration network proposed in:
Hu et al., Label-driven weakly-supervised learning for
multimodal deformable image registration, arXiv:1711.01666
https://arxiv.org/abs/1711.01666
Hu et al., Weakly-Supervised Convolutional Neural Networks for
Multimodal Image Registration, Medical Image Analysis (2018)
https://doi.org/10.1016/j.media.2018.07.002
see also:
https://github.com/YipengHu/label-reg
:param decay:
:param affine_w_initializer:
:param affine_b_initializer:
:param disp_w_initializer:
:param disp_b_initializer:
:param acti_func:
:param interp:
:param boundary:
:param name:
"""
BaseNet.__init__(self, name=name)
self.global_net = INetAffine(decay=decay,
affine_w_initializer=affine_w_initializer,
affine_b_initializer=affine_b_initializer,
acti_func=acti_func,
name='inet-global')
self.local_net = INetDense(decay=decay,
disp_w_initializer=disp_w_initializer,
disp_b_initializer=disp_b_initializer,
acti_func=acti_func,
name='inet-local')
self.interp = interp
self.boundary = boundary
def __init__(self,
decay=1e-6,
affine_w_initializer=None,
affine_b_initializer=None,
acti_func='relu',
name='inet-affine'):
"""
This network estimates affine transformations from
a pair of moving and fixed image:
Hu et al., Label-driven weakly-supervised learning for
multimodal deformable image registration, arXiv:1711.01666
https://arxiv.org/abs/1711.01666
Hu et al., Weakly-Supervised Convolutional Neural Networks for
Multimodal Image Registration, Medical Image Analysis (2018)
https://doi.org/10.1016/j.media.2018.07.002
see also:
https://github.com/YipengHu/label-reg
:param decay:
:param affine_w_initializer:
:param affine_b_initializer:
:param acti_func:
:param name:
"""
BaseNet.__init__(self, name=name)
self.fea = [4, 8, 16, 32, 64]
self.k_conv = 3
self.affine_w_initializer = affine_w_initializer
self.affine_b_initializer = affine_b_initializer
self.res_param = {
'w_initializer': GlorotUniform.get_instance(''),
'w_regularizer': regularizers.l2_regularizer(decay),
'acti_func': acti_func}
self.affine_param = {
'w_regularizer': regularizers.l2_regularizer(decay),
'b_regularizer': None}
def __init__(self,
decay,
affine_w_initializer=None,
affine_b_initializer=None,
disp_w_initializer=None,
disp_b_initializer=None,
acti_func='relu',
interp='linear',
boundary='replicate',
name='inet-hybrid-pre-warp'):
"""
Re-implementation of the registration network proposed in:
Hu et al., Label-driven weakly-supervised learning for
multimodal deformable image registration, arXiv:1711.01666
https://arxiv.org/abs/1711.01666
:param decay:
:param affine_w_initializer:
:param affine_b_initializer:
:param disp_w_initializer:
:param disp_b_initializer:
:param acti_func:
:param interp:
:param boundary:
:param name:
"""
BaseNet.__init__(self, name=name)
self.global_net = INetAffine(decay=decay,
affine_w_initializer=affine_w_initializer,
affine_b_initializer=affine_b_initializer,
acti_func=acti_func,
name='inet-global')
self.local_net = INetDense(decay=decay,
disp_w_initializer=disp_w_initializer,
disp_b_initializer=disp_b_initializer,
acti_func=acti_func,
name='inet-local')
self.interp = interp
self.boundary = boundary
def __init__(self,
num_classes,
w_initializer=None,
w_regularizer=None,
b_initializer=None,
b_regularizer=None,
acti_func='leakyrelu',
name='naive_net'):
BaseNet.__init__(self, num_classes=num_classes, name=name)
self.n_fea = [512, 512, 2]
net_params = {
'padding': 'SAME',
'with_bias': True,
'with_bn': False,
'acti_func': acti_func,
'w_initializer': w_initializer,
'b_initializer': b_initializer,
'w_regularizer': w_regularizer,
'b_regularizer': b_regularizer
}
def __init__(self,
decay,
affine_w_initializer=None,
affine_b_initializer=None,
disp_w_initializer=None,
disp_b_initializer=None,
acti_func='relu',
interp='linear',
boundary='replicate',
name='inet-hybrid-two-stream'):
BaseNet.__init__(self, name=name)
self.global_net = INetAffine(decay=decay,
affine_w_initializer=affine_w_initializer,
affine_b_initializer=affine_b_initializer,
acti_func=acti_func,
name='inet-global')
self.local_net = INetDense(decay=decay,
disp_w_initializer=disp_w_initializer,
disp_b_initializer=disp_b_initializer,
acti_func=acti_func,
name='inet-local')
self.interp = interp
self.boundary = boundary
def __init__(self,
decay=1e-6,
affine_w_initializer=None,
affine_b_initializer=None,
acti_func='relu',
name='inet-affine'):
"""
This network estimates affine transformations from
a pair of moving and fixed image:
Hu et al., Label-driven weakly-supervised learning for
multimodal deformable image registration, arXiv:1711.01666
https://arxiv.org/abs/1711.01666
:param decay:
:param affine_w_initializer:
:param affine_b_initializer:
:param acti_func:
:param name:
"""
BaseNet.__init__(self, name=name)
self.fea = [4, 8, 16, 32, 64]
self.k_conv = 3
self.affine_w_initializer = affine_w_initializer
self.affine_b_initializer = affine_b_initializer
self.res_param = {
'w_initializer': GlorotUniform.get_instance(''),
'w_regularizer': regularizers.l2_regularizer(decay),
'acti_func': acti_func
}
self.affine_param = {
'w_regularizer': regularizers.l2_regularizer(decay),
'b_regularizer': None
}
def __init__(self):
BaseNet.__init__(self, name='G')
def __init__(self,
decay=0.0,
smoothing=0,
disp_w_initializer=None,
disp_b_initializer=None,
acti_func='relu',
multi_scale_fusion=True,
name='inet-dense'):
"""
The network estimates dense displacement fields from a pair
of moving and fixed images:
Hu et al., Label-driven weakly-supervised learning for
multimodal deformable image registration, arXiv:1711.01666
https://arxiv.org/abs/1711.01666
Hu et al., Weakly-Supervised Convolutional Neural Networks for
Multimodal Image Registration, Medical Image Analysis (2018)
https://doi.org/10.1016/j.media.2018.07.002
see also:
https://github.com/YipengHu/label-reg
:param decay:
:param smoothing:
:param disp_w_initializer: initialisation of the displacement fields
:param disp_b_initializer: initialisation of the dis
:param acti_func:
:param multi_scale_fusion: True/False indicating whether to use
multiscale feature fusion.
:param name:
"""
BaseNet.__init__(self, name=name)
# self.fea = [40, 80, 160, 320, 640]
# self.fea = [32, 64, 128, 256, 512]
self.fea = [30, 60, 120, 240, 480]
# self.fea = [16, 32, 64, 128, 256]
self.k_conv = 3
self.multi_scale_fusion = multi_scale_fusion
self.down_res_param = {
'w_initializer': GlorotUniform.get_instance(''),
'w_regularizer': regularizers.l2_regularizer(decay),
'acti_func': acti_func
}
self.up_res_param = {
'acti_func': acti_func,
'w_initializer': GlorotUniform.get_instance(''),
'w_regularizer': regularizers.l2_regularizer(decay),
'is_residual_upsampling': True,
'type_string': 'bn_acti_conv'
}
# displacement initialiser & regulariser
if disp_w_initializer is None:
disp_b_initializer = tf.constant_initializer(0.0)
#disp_w_initializer = tf.random_normal_initializer(0, 1e-4)
if disp_b_initializer is None:
disp_b_initializer = tf.constant_initializer(0.0)
#disp_w_initializer = tf.random_normal_initializer(0, 0.0)
self.disp_param = {
'w_initializer': disp_w_initializer,
'w_regularizer': regularizers.l2_regularizer(decay),
'b_initializer': disp_b_initializer,
'b_regularizer': None
}
if smoothing > 0:
self.smoothing_func = _smoothing_func(smoothing)
else:
self.smoothing_func = None
def __init__(self):
BaseNet.__init__(self, name='G')
def __init__(self):
BaseNet.__init__(self, name='tinynet')
self.d_net = DNet()
self.g_net = GNet()
def __init__(self,
decay=0.0,
smoothing=0,
disp_w_initializer=None,
disp_b_initializer=None,
acti_func='relu',
multi_scale_fusion=True,
name='inet-dense'):
"""
The network estimates dense displacement fields from a pair
of moving and fixed images:
Hu et al., Label-driven weakly-supervised learning for
multimodal deformable image registration, arXiv:1711.01666
https://arxiv.org/abs/1711.01666
:param decay:
:param smoothing:
:param disp_w_initializer: initialisation of the displacement fields
:param disp_b_initializer: initialisation of the dis
:param acti_func:
:param multi_scale_fusion: True/False indicating whether to use
multiscale feature fusion.
:param name:
"""
BaseNet.__init__(self, name=name)
# self.fea = [40, 80, 160, 320, 640]
# self.fea = [32, 64, 128, 256, 512]
self.fea = [30, 60, 120, 240, 480]
# self.fea = [16, 32, 64, 128, 256]
self.k_conv = 3
self.multi_scale_fusion = multi_scale_fusion
self.down_res_param = {
'w_initializer': GlorotUniform.get_instance(''),
'w_regularizer': regularizers.l2_regularizer(decay),
'acti_func': acti_func}
self.up_res_param = {
'acti_func': acti_func,
'w_initializer': GlorotUniform.get_instance(''),
'w_regularizer': regularizers.l2_regularizer(decay),
'is_residual_upsampling': True,
'type_string': 'bn_acti_conv'}
# displacement initialiser & regulariser
if disp_w_initializer is None:
disp_b_initializer = tf.constant_initializer(0.0)
#disp_w_initializer = tf.random_normal_initializer(0, 1e-4)
if disp_b_initializer is None:
disp_b_initializer = tf.constant_initializer(0.0)
#disp_w_initializer = tf.random_normal_initializer(0, 0.0)
self.disp_param = {
'w_initializer': disp_w_initializer,
'w_regularizer': regularizers.l2_regularizer(decay),
'b_initializer': disp_b_initializer,
'b_regularizer': None}
if smoothing > 0:
self.smoothing_func = _smoothing_func(smoothing)
else:
self.smoothing_func = None
def __init__(self):
BaseNet.__init__(self, name='tinynet')
self.d_net = DNet()
self.g_net = GNet()