def __init__(self, kwargs):
Layer.__init__(self, kwargs)
self.name = ModelAssign(kwargs, 'name', None)
self.input = ModelAssign(kwargs, 'input', None)
self.filters = ModelAssign(kwargs, 'filters', 32)
self.kernel_size = ModelAssign(kwargs, 'kernel_size', 3)
self.strides = ModelAssign(kwargs, 'strides', 1)
self.padding = ModelAssign(kwargs, 'padding', 'SAME')
self.hyper = ModelAssign(kwargs, 'hyper', False)
self.batchnorm = ModelAssign(kwargs, 'batchnorm', False)
self.activation = ModelAssign(kwargs, 'activation', 'relu')
self.use_bias = ModelAssign(kwargs, 'use_bias', True)
self.num_trainable_parameters = 0
self.num_non_trainable_parameters = 0
self.shared_trainable_parameters = 0
self.skip_from_names = None
self.MACs = 0
self.peak_activation_mem = 0
self.dropout = ModelAssign(kwargs, 'dropout', 0.0)
self.dropout_tensor = None
# HyperNetwork parameters
if not self.hyper:
self.zdims = None
self.layer_info = None
self.basic_block_size = None
self.hidden = None
else:
raise NotImplementedError
self.input_shape = None
self.output_shape = None
self.output_tensor = None
def __init__(self, kwargs):
"""
Accuracy.
:param kwargs:
"""
Layer.__init__(self, kwargs)
self.name = ModelAssign(kwargs, 'name', None)
self.logits = ModelAssign(kwargs, 'logits', None)
self.labels = ModelAssign(kwargs, 'labels', None)
self.output_tensor = None
def __init__(self, kwargs):
Layer.__init__(self, kwargs)
self.name = ModelAssign(kwargs, 'name', None)
self.input_shape = None
self.output_shape = None
self.output_tensor = None
self.num_trainable_parameters = 0
self.num_non_trainable_parameters = 0
self.shared_trainable_parameters = 0
self.MACs = 0
def __init__(self, kwargs):
"""
Top-k accuracy for large-scale image classification. (ImageNet)
:param kwargs: Configurations
"""
Layer.__init__(self, kwargs)
self.name = ModelAssign(kwargs, 'name', None)
self.logits = ModelAssign(kwargs, 'logits', None)
self.labels = ModelAssign(kwargs, 'labels', None)
self.k = ModelAssign(kwargs, 'k', 5)
self.output_tensor = None
def __init__(self, kwargs):
Layer.__init__(self, kwargs)
self.name = ModelAssign(kwargs, 'name', None)
self.input_shape = Str2List(ModelAssign(kwargs, 'input_shape', None))
self.dtype = ModelAssign(kwargs, 'dtype', 'float32')
self.output_shape = None
self.output_tensor = None
self.mean = ModelAssign(kwargs, 'mean', None)
self.std = ModelAssign(kwargs, 'std', None)
self.num_trainable_parameters = 0
self.num_non_trainable_parameters = 0
self.shared_trainable_parameters = 0
self.MACs = 0
def __init__(self, kwargs):
Layer.__init__(self, kwargs)
self.name = ModelAssign(kwargs, 'name', None)
self.axis = ModelAssign(kwargs, 'axis', -1)
self.activation = ModelAssign(kwargs, 'activation', None)
self.input_shape = None
self.output_shape = None
self.output_tensor = None
self.num_trainable_parameters = 0
self.num_non_trainable_parameters = 0
self.shared_trainable_parameters = 0
self.MACs = 0
self.peak_activation_mem = 0
def __init__(self, kwargs):
"""
Concatenation. Concatenate two tensors if possible.
:param kwargs: configurations for concatenation layers.
"""
Layer.__init__(self, kwargs)
self.name = ModelAssign(kwargs, 'name', None)
self.activation = ModelAssign(kwargs, 'activation', None)
self.input_shape = None
self.output_shape = None
self.output_tensor = None
self.num_trainable_parameters = 0
self.num_non_trainable_parameters = 0
self.shared_trainable_parameters = 0
self.mem_cost = 0
self.MACs = 0
self.peak_activation_mem = 0
def __init__(self, kwargs):
Layer.__init__(self, kwargs)
self.name = ModelAssign(kwargs, 'name', None)
self.input = ModelAssign(kwargs, 'input', None)
self.kernel_size = ModelAssign(kwargs, 'kernel_size', 3)
self.strides = ModelAssign(kwargs, 'strides', 1)
self.padding = ModelAssign(kwargs, 'padding', 'SAME')
self.hyper = ModelAssign(kwargs, 'hyper', False)
self.batchnorm = ModelAssign(kwargs, 'batchnorm', False)
self.activation = ModelAssign(kwargs, 'activation', 'relu')
self.use_bias = ModelAssign(kwargs, 'use_bias', True)
self.depthwise_multiplier = ModelAssign(kwargs, 'depthwise_multiplier',
1)
self.num_trainable_parameters = 0
self.num_non_trainable_parameters = 0
self.shared_trainable_parameters = 0
self.mem_cost = 0
self.skip_from_names = None
self.MACs = 0
self.peak_activation_mem = 0
# HyperNetwork parameters
if not self.hyper:
self.zdims = None
self.layer_info = None
self.basic_block_size = None
self.hidden = None
else:
self.zdims = ModelAssign(kwargs, 'hyper_zdims', 4)
self.layer_info = tf.placeholder(dtype=tf.float32,
shape=[1, self.zdims],
name=self.name + 'layer_info')
self.basic_block_size = Str2List(
ModelAssign(kwargs, 'hyper_basic_block_size', None))
self.hidden = ModelAssign(kwargs, 'hyper_hidden', 16)
self.input_shape = None
self.output_shape = None
self.output_tensor = None
def __init__(self, kwargs):
Layer.__init__(self, kwargs)
self.name = ModelAssign(kwargs, 'name', None)
self.input = ModelAssign(kwargs, 'input', None)
self.units = ModelAssign(kwargs, 'units', 10)
self.use_bias = ModelAssign(kwargs, 'use_bias', True)
self.batchnorm = ModelAssign(kwargs, 'batchnorm', False)
self.trainable = ModelAssign(kwargs, 'trainable', True)
self.activation = ModelAssign(kwargs, 'activation', 'relu')
self.dropout = ModelAssign(kwargs, 'dropout', 0.0)
self.skip_from_names = None
self.input_shape = None
self.output_shape = None
self.output_tensor = None
self.dropout_tensor = None
# Params
self.num_trainable_parameters = 0
self.num_non_trainable_parameters = 0
self.shared_trainable_parameters = 0
self.MACs = 0
self.mem_cost = 0
self.peak_activation_mem = 0
def __init__(self, kwargs):
Layer.__init__(self, kwargs)
self.name = ModelAssign(kwargs, 'name', None)
self.output_tensor = None
def __init__(self, kwargs):
Layer.__init__(self, kwargs)
pass
