def __call__(self, shape, dtype="float32"):
""" Call function for the ICNR initializer.
Parameters
----------
shape: tuple or list
The required resized shape for the output tensor
dtype: str
The data type for the tensor
Returns
-------
tensor
The modified kernel weights
"""
shape = list(shape)
if self.scale == 1:
return self.initializer(shape)
new_shape = shape[:3] + [shape[3] // (self.scale**2)]
if isinstance(self.initializer, dict):
self.initializer = initializers.deserialize(self.initializer)
var_x = self.initializer(new_shape, dtype)
var_x = K.permute_dimensions(var_x, [2, 0, 1, 3])
var_x = self._resize_nearest_neighbour(
var_x, (shape[0] * self.scale, shape[1] * self.scale))
var_x = self._space_to_depth(var_x)
var_x = K.permute_dimensions(var_x, [1, 2, 0, 3])
logger.debug("Output: %s", var_x)
return var_x
def test_serialization_deserialization(self, cls_name, expected_cls):
initializer = initializers.get(cls_name)
self.assertIsInstance(initializer, expected_cls)
config = initializers.serialize(initializer)
recreated = initializers.deserialize(config)
self.assertIsInstance(recreated, expected_cls)
self.assertEqual(config, initializers.serialize(recreated))
def test_load_external_variance_scaling_v2(self):
external_serialized_json = {
'class_name': 'VarianceScaling',
'config': {
'distribution': 'normal',
'mode': 'fan_avg',
'scale': 1.0,
'seed': None
}
}
initializer = initializers.deserialize(external_serialized_json)
self.assertEqual(initializer.distribution, 'truncated_normal')
def test_load_external_variance_scaling_v2(self):
external_serialized_json = {
"class_name": "VarianceScaling",
"config": {
"distribution": "normal",
"mode": "fan_avg",
"scale": 1.0,
"seed": None,
},
}
initializer = initializers.deserialize(external_serialized_json)
self.assertEqual(initializer.distribution, "truncated_normal")
def __call__(self, shape, dtype='float32'): # tf needs partition_info=None
shape = list(shape)
if self.scale == 1:
return self.initializer(shape)
new_shape = shape[:3] + [shape[3] // (self.scale ** 2)]
if type(self.initializer) is dict:
self.initializer = initializers.deserialize(self.initializer)
var_x = self.initializer(new_shape, dtype)
var_x = tf.transpose(var_x, perm=[2, 0, 1, 3])
var_x = tf.image.resize_nearest_neighbor(
var_x,
size=(shape[0] * self.scale, shape[1] * self.scale),
align_corners=True)
var_x = tf.space_to_depth(var_x, block_size=self.scale, data_format='NHWC')
var_x = tf.transpose(var_x, perm=[1, 2, 0, 3])
return var_x
def __call__(self, shape, dtype='float32'): # tf needs partition_info=None
shape = list(shape)
if self.scale == 1:
return self.initializer(shape)
new_shape = shape[:3] + [shape[3] // (self.scale ** 2)]
if isinstance(self.initializer, dict):
# 反序列化
self.initializer = initializers.deserialize(self.initializer)
var_x = self.initializer(new_shape, dtype)
var_x = tf.transpose(var_x, perm=[2, 0, 1, 3])
# 使用最近邻插值调整images为size，输入和输出张量的4个角像素的中心对齐,并保留角落像素处的值
var_x = tf.image.resize_nearest_neighbor(
var_x,
size=(shape[0] * self.scale, shape[1] * self.scale),
align_corners=True)
var_x = tf.space_to_depth(var_x, block_size=self.scale, data_format='NHWC')
var_x = tf.transpose(var_x, perm=[1, 2, 0, 3])
return var_x
