hub / github.com/LynnHo/EigenGAN-Tensorflow / convolution

Function convolution

tflib/deep_learning/layers.py:67–129 · view source on GitHub ↗

(inputs,
                num_outputs,
                kernel_size,
                stride=1,
                padding='SAME',
                data_format=None,
                rate=1,
                activation_fn=None,
                normalizer_fn=None,
                normalizer_params=None,
                weights_normalizer_fn=None,
                weights_normalizer_params=None,
                weights_initializer=tf.glorot_uniform_initializer(),
                weights_regularizer=None,
                biases_initializer=tf.zeros_initializer(),
                biases_regularizer=None,
                reuse=None,
                trainable=True,
                scope=None)

Source from the content-addressed store, hash-verified

65
66
67	def convolution(inputs,
68	num_outputs,
69	kernel_size,
70	stride=1,
71	padding='SAME',
72	data_format=None,
73	rate=1,
74	activation_fn=None,
75	normalizer_fn=None,
76	normalizer_params=None,
77	weights_normalizer_fn=None,
78	weights_normalizer_params=None,
79	weights_initializer=tf.glorot_uniform_initializer(),
80	weights_regularizer=None,
81	biases_initializer=tf.zeros_initializer(),
82	biases_regularizer=None,
83	reuse=None,
84	trainable=True,
85	scope=None):
86	with tf.variable_scope(scope, 'convolution', reuse=reuse):
87	conv_dims = inputs.shape.rank - 2
88	kernel_size = kernel_size if isinstance(kernel_size, (list, tuple)) else [kernel_size] * conv_dims
89	stride = stride if isinstance(stride, (list, tuple)) else [stride] * conv_dims
90	rate = rate if isinstance(rate, (list, tuple)) else [rate] * conv_dims
91	if data_format is None or data_format.endswith('C'):
92	num_inputs = inputs.shape[-1]
93	elif data_format.startswith('NC'):
94	num_inputs = inputs.shape[1]
95	else:
96	raise ValueError('Invalid data_format')
97
98	weights = tf.get_variable('weights',
99	shape=list(kernel_size) + [num_inputs, num_outputs],
100	initializer=weights_initializer,
101	regularizer=weights_regularizer,
102	trainable=trainable)
103	if weights_normalizer_fn is not None:
104	weights_normalizer_params = weights_normalizer_params or {}
105	weights = weights_normalizer_fn(weights, **weights_normalizer_params)
106
107	outputs = tf.nn.convolution(input=inputs,
108	filter=weights,
109	dilation_rate=rate,
110	strides=stride,
111	padding=padding,
112	data_format=data_format)
113
114	if normalizer_fn is not None:
115	normalizer_params = normalizer_params or {}
116	outputs = normalizer_fn(outputs, **normalizer_params)
117	else:
118	if biases_initializer is not None:
119	biases = tf.get_variable('biases',
120	shape=[num_outputs],
121	initializer=biases_initializer,
122	regularizer=biases_regularizer,
123	trainable=trainable)
124	outputs = tf.nn.bias_add(outputs, biases, data_format=data_format)

Callers

nothing calls this directly

Calls 1

activation_fnFunction · 0.85

Tested by

no test coverage detected