def __new__(cls, x, w, b, filter=3, stride=1, padding=0):
filter, stride, padding = (tuplize(x)
for x in (filter, stride, padding))
in_shape = x.shape[1:]
out_shape = [w.shape[0]]
out_shape.extend(out_size(x.shape[2:], filter, stride, padding))
return cls.calc_value(x, w, b, in_shape, out_shape, filter, stride,
padding)
def __new__(cls, x, index, filter, stride, padding):
filter, stride, padding = (tuplize(x)
for x in (filter, stride, padding))
in_shape = x.shape[1:]
out_shape = [
x.shape[1],
]
out_shape.extend(
transpose_out_size(in_shape[1:], filter, stride, padding))
return cls.calc_value(x, index, in_shape, out_shape, filter, stride,
padding)
def __new__(cls, x, filter=3, stride=1, padding=0, ceil_mode=False):
filter, stride, padding = (tuplize(x)
for x in (filter, stride, padding))
in_shape = x.shape[1:]
out_shape = [
x.shape[1],
]
out_shape.extend(
out_size(x.shape[2:], filter, stride, padding,
ceil_mode=ceil_mode))
return cls.calc_value(x, in_shape, out_shape, filter, stride, padding)
def __init__(self,
channel=32,
filter=3,
padding=0,
stride=1,
input_size=None,
initializer=GlorotNormal()):
self._padding, self._stride, self._kernel = (tuplize(x)
for x in (padding, stride,
filter))
self._channel = channel
self._initializer = initializer
super(Conv2d, self).__init__(input_size)
def __new__(cls, x, w, b, filter=3, stride=1, padding=0, dilation=1):
filter, stride, padding, dilation = (tuplize(x)
for x in (filter, stride, padding,
dilation))
in_shape = x.shape[1:]
out_shape = [
w.shape[1],
]
out_shape.extend(
transpose_out_size(in_shape[1:], filter, stride, padding,
dilation))
return cls.calc_value(x, w, b, in_shape, out_shape, filter, stride,
padding, dilation)
def __init__(self,
channel=32,
filter=3,
padding=0,
stride=1,
dilation=1,
input_size=None,
ignore_bias=False,
initializer=GlorotNormal(),
weight_decay=0):
self._padding, self._stride, self._kernel, self._dilation = (
tuplize(x) for x in (padding, stride, filter, dilation))
self._channel = channel
self._ignore_bias = ignore_bias
self._initializer = initializer
self._weight_decay = weight_decay
super(Conv2d, self).__init__(input_size)
def __init__(self,
channel=32,
filter=3,
padding=0,
stride=1,
dilation=1,
groups=1,
input_size=None,
ignore_bias=False,
initializer=GlorotNormal(),
weight_decay=0):
self._padding, self._stride, self._kernel, self._dilation = (
tuplize(x) for x in (padding, stride, filter, dilation))
self._channel = channel
self._groups = groups
#print("self._groups: ", self._groups)
assert isinstance(
self._groups, int
) and self._groups > 0, "Please set groups to integer greater than 0"
self._ignore_bias = ignore_bias
self._initializer = initializer
self._weight_decay = weight_decay
super(GroupConv2d, self).__init__(input_size)
def __init__(self, filter=3, padding=0, stride=1, ceil_mode=False):
self._padding, self._stride, self._kernel = (tuplize(x)
for x in (padding, stride,
filter))
self._ceil_mode = ceil_mode
def __init__(self, filter=3,
padding=0, stride=1):
self._padding, self._stride, self._kernel = (tuplize(x) for x in (padding, stride, filter))
