import math
import torch
from torch.nn.parameter import Parameter
import torch.nn.functional as F
import torch.nn as nn
from torch.nn.modules.utils import _single, _pair, _triple
from .data_wrapper import MyTensor, USE_CUDA
device = torch.device("cuda:0" if (USE_CUDA and torch.cuda.is_available()) else "cpu")
[docs]class NoisyLinear(nn.Module):
"""Applies a noisy linear transformation to the incoming data: :math:`y = (mu_w + sigma_w \cdot epsilon_w)x + mu_b + sigma_b \cdot epsilon_b`
More details can be found in the paper `ZZ` _ .
Args:
in_features: size of each input sample
out_features: size of each output sample
bias: If set to False, the layer will not learn an additive bias. Default: True
factorised: whether or not to use factorised noise. Default: True
std_init: initialization constant for standard deviation component of weights. If None, \
defaults to 0.017 for independent and 0.4 for factorised. Default: None
Shape:
- Input: (N, in_features)
- Output:(N, out_features)
Attributes:
weight: the learnable weights of the module of shape (out_features x in_features)
bias: the learnable bias of the module of shape (out_features)
Examples::
>>> m = nn.NoisyLinear(20, 30)
>>> input = autograd.Variable(torch.randn(128, 20))
>>> output = m(input)
>>> print(output.size())
"""
def __init__(self, in_features, out_features, bias=True, factorised=True, std_init=None):
super(NoisyLinear, self).__init__()
self.in_features = in_features
self.out_features = out_features
self.factorised = factorised
self.weight_mu = Parameter(MyTensor(out_features, in_features))
self.weight_sigma = Parameter(MyTensor(out_features, in_features))
if bias:
self.bias_mu = Parameter(MyTensor(out_features))
self.bias_sigma = Parameter(MyTensor(out_features))
else:
self.register_parameter('bias', None)
if std_init is None:
if self.factorised:
self.std_init = 0.4
else:
self.std_init = 0.017
else:
self.std_init = std_init
self.reset_parameters(bias)
[docs] def reset_parameters(self, bias):
if self.factorised:
mu_range = 1. / math.sqrt(self.weight_mu.size(1))
self.weight_mu.data.uniform_(-mu_range, mu_range)
self.weight_sigma.data.fill_(self.std_init / math.sqrt(self.weight_sigma.size(1)))
if bias:
self.bias_mu.data.uniform_(-mu_range, mu_range)
self.bias_sigma.data.fill_(self.std_init / math.sqrt(self.bias_sigma.size(0)))
else:
mu_range = math.sqrt(3. / self.weight_mu.size(1))
self.weight_mu.data.uniform_(-mu_range, mu_range)
self.weight_sigma.data.fill_(self.std_init)
if bias:
self.bias_mu.data.uniform_(-mu_range, mu_range)
self.bias_sigma.data.fill_(self.std_init)
[docs] def scale_noise(self, size):
x = MyTensor(size).normal_()
x = x.sign().mul(x.abs().sqrt())
return x
[docs] def forward(self, input):
if self.factorised:
epsilon_in = self.scale_noise(self.in_features)
epsilon_out = self.scale_noise(self.out_features)
weight_epsilon = epsilon_out.ger(epsilon_in)
bias_epsilon = self.scale_noise(self.out_features)
else:
weight_epsilon = MyTensor(*(self.out_features, self.in_features)).normal_()
bias_epsilon = MyTensor(self.out_features).normal_()
return F.linear(input,
self.weight_mu + self.weight_sigma.mul(weight_epsilon),
self.bias_mu + self.bias_sigma.mul(bias_epsilon))
def __repr__(self):
return self.__class__.__name__ + ' (' \
+ str(self.in_features) + ' -> ' \
+ str(self.out_features) + ')'
[docs]class NoisyLayer(nn.Module):
def __init__(self, std_init=None, start_reducing_from_iter=25):
super(NoisyLayer, self).__init__()
self.std_init = std_init
if self.std_init is None:
self.std_init = 0.25
else:
self.std_init = std_init
self.start_reducing_from_iter = start_reducing_from_iter
[docs] def forward(self, input, iter=0):
noise_epsilon = MyTensor(input.size()).normal_()
if self.training:
effective_iter = max(0,iter-self.start_reducing_from_iter)
output = input + 1. / (effective_iter + 1) * self.std_init * noise_epsilon
else:
output = input
return output
class _NoisyConvNd(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size, stride,
padding, dilation, transposed, output_padding, groups, bias, scalar_sigmas=True, optimize_sigmas=False, std_init=None, start_reducing_from_iter=25):
super(_NoisyConvNd, self).__init__()
if in_channels % groups != 0:
raise ValueError('in_channels must be divisible by groups')
if out_channels % groups != 0:
raise ValueError('out_channels must be divisible by groups')
self.in_channels = in_channels
self.out_channels = out_channels
self.kernel_size = kernel_size
self.stride = stride
self.padding = padding
self.dilation = dilation
self.transposed = transposed
self.output_padding = output_padding
self.groups = groups
self.scalar_sigmas = scalar_sigmas
self.optimize_sigmas = optimize_sigmas
self.std_init = std_init
self.start_reducing_from_iter = start_reducing_from_iter
if self.std_init is None:
self.std_init = 0.25
else:
self.std_init = std_init
if transposed:
self.weight = Parameter(MyTensor(
in_channels, out_channels // groups, *kernel_size))
if self.scalar_sigmas:
if self.optimize_sigmas:
self.weight_sigma = Parameter(MyTensor(1))
else:
self.weight_sigma = MyTensor(1)
else:
if self.optimize_sigmas:
self.weight_sigma = Parameter(MyTensor(in_channels, out_channels//groups))
else:
self.weight_sigma = MyTensor(in_channels, out_channels//groups)
else:
self.weight = Parameter(MyTensor(
out_channels, in_channels // groups, *kernel_size))
if self.scalar_sigmas:
if self.optimize_sigmas:
self.weight_sigma = Parameter(MyTensor(1))
else:
self.weight_sigma = MyTensor(1)
else:
if self.optimize_sigmas:
self.weight_sigma = Parameter(MyTensor(out_channels, in_channels//groups))
else:
self.weight_sigma = MyTensor(out_channels, in_channels // groups)
if bias:
self.bias = Parameter(MyTensor(out_channels))
if self.scalar_sigmas:
if self.optimize_sigmas:
self.bias_sigma = Parameter(MyTensor(1))
else:
self.bias_sigma = MyTensor(1)
else:
if self.optimize_sigmas:
self.bias_sigma = Parameter(MyTensor(out_channels))
else:
self.bias_sigma = MyTensor(out_channels)
else:
self.register_parameter('bias', None)
self.register_parameter('bias_sigma', None)
self.reset_parameters(bias)
def reset_parameters(self, bias):
# todo: adapt this to the used type of nonlinearity
nn.init.kaiming_normal_(self.weight.data)
self.weight_sigma.data.fill_(self.std_init)
if bias:
self.bias.data.fill_(0)
self.bias_sigma.data.fill_(self.std_init)
#mu_range = math.sqrt(3. / self.weight.size(1))
#self.weight.data.uniform_(-mu_range, mu_range)
#self.weight_sigma.data.fill_(self.std_init)
#if bias:
# self.bias.data.uniform_(-mu_range, mu_range)
# self.bias_sigma.data.fill_(self.std_init)
#n = self.in_channels
#for k in self.kernel_size:
# n *= k
#stdv = 1. / math.sqrt(n)
#self.weight.data.uniform_(-stdv, stdv)
#if self.bias is not None:
# self.bias.data.uniform_(-stdv, stdv)
def extra_repr(self):
s = ('{in_channels}, {out_channels}, kernel_size={kernel_size}'
', stride={stride}')
if self.padding != (0,) * len(self.padding):
s += ', padding={padding}'
if self.dilation != (1,) * len(self.dilation):
s += ', dilation={dilation}'
if self.output_padding != (0,) * len(self.output_padding):
s += ', output_padding={output_padding}'
if self.groups != 1:
s += ', groups={groups}'
if self.bias is None:
s += ', bias=False'
return s.format(**self.__dict__)
[docs]class NoisyConv1d(_NoisyConvNd):
r"""Applies a 1D noisy convolution over an input signal composed of several input
planes.
In the simplest case, the output value of the layer with input size
:math:`(N, C_{in}, L)` and output :math:`(N, C_{out}, L_{out})` can be
precisely described as:
.. math::
\begin{equation*}
\text{out}(N_i, C_{out_j}) = \text{bias}(C_{out_j}) +
\sum_{k = 0}^{C_{in} - 1} \text{weight}(C_{out_j}, k) \star \text{input}(N_i, k)
\end{equation*},
where :math:`\star` is the valid `cross-correlation`_ operator,
:math:`N` is a batch size, :math:`C` denotes a number of channels,
:math:`L` is a length of signal sequence.
* :attr:`stride` controls the stride for the cross-correlation, a single
number or a one-element tuple.
* :attr:`padding` controls the amount of implicit zero-paddings on both sides
for :attr:`padding` number of points.
* :attr:`dilation` controls the spacing between the kernel points; also
known as the à trous algorithm. It is harder to describe, but this `link`_
has a nice visualization of what :attr:`dilation` does.
* :attr:`groups` controls the connections between inputs and outputs.
:attr:`in_channels` and :attr:`out_channels` must both be divisible by
:attr:`groups`. For example,
* At groups=1, all inputs are convolved to all outputs.
* At groups=2, the operation becomes equivalent to having two conv
layers side by side, each seeing half the input channels,
and producing half the output channels, and both subsequently
concatenated.
* At groups= :attr:`in_channels`, each input channel is convolved with
its own set of filters (of size
:math:`\left\lfloor \frac{\text{out_channels}}{\text{in_channels}} \right\rfloor`).
.. note::
Depending of the size of your kernel, several (of the last)
columns of the input might be lost, because it is a valid
`cross-correlation`_, and not a full `cross-correlation`_.
It is up to the user to add proper padding.
.. note::
The configuration when `groups == in_channels` and `out_channels == K * in_channels`
where `K` is a positive integer is termed in literature as depthwise convolution.
In other words, for an input of size :math:`(N, C_{in}, L_{in})`, if you want a
depthwise convolution with a depthwise multiplier `K`,
then you use the constructor arguments
:math:`(\text{in_channels}=C_{in}, \text{out_channels}=C_{in} * K, ..., \text{groups}=C_{in})`
Args:
in_channels (int): Number of channels in the input image
out_channels (int): Number of channels produced by the convolution
kernel_size (int or tuple): Size of the convolving kernel
stride (int or tuple, optional): Stride of the convolution. Default: 1
padding (int or tuple, optional): Zero-padding added to both sides of
the input. Default: 0
dilation (int or tuple, optional): Spacing between kernel
elements. Default: 1
groups (int, optional): Number of blocked connections from input
channels to output channels. Default: 1
bias (bool, optional): If ``True``, adds a learnable bias to the output. Default: ``True``
Shape:
- Input: :math:`(N, C_{in}, L_{in})`
- Output: :math:`(N, C_{out}, L_{out})` where
.. math::
L_{out} = \left\lfloor\frac{L_{in} + 2 \times \text{padding} - \text{dilation}
\times (\text{kernel_size} - 1) - 1}{\text{stride}} + 1\right\rfloor
Attributes:
weight (Tensor): the learnable weights of the module of shape
(out_channels, in_channels, kernel_size)
bias (Tensor): the learnable bias of the module of shape
(out_channels)
Examples::
>>> m = nn.NoisyConv1d(16, 33, 3, stride=2)
>>> input = torch.randn(20, 16, 50)
>>> output = m(input)
.. _cross-correlation:
https://en.wikipedia.org/wiki/Cross-correlation
.. _link:
https://github.com/vdumoulin/conv_arithmetic/blob/master/README.md
"""
def __init__(self, in_channels, out_channels, kernel_size, stride=1,
padding=0, dilation=1, groups=1, bias=True,
scalar_sigmas=True, optimize_sigmas=False, std_init=None, start_reducing_from_iter=25):
kernel_size = _single(kernel_size)
stride = _single(stride)
padding = _single(padding)
dilation = _single(dilation)
super(NoisyConv1d, self).__init__(
in_channels, out_channels, kernel_size, stride, padding, dilation,
False, _single(0), groups, bias, scalar_sigmas, optimize_sigmas, std_init, start_reducing_from_iter)
[docs] def forward(self, input, iter=0):
weight_epsilon = MyTensor(*(self.out_channels, self.in_channels, *self.kernel_size)).normal_()
bias_epsilon = MyTensor(self.out_channels).normal_()
if self.bias is not None:
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_bias_value = self.bias + 1./(effective_iter+1)*self.bias_sigma * bias_epsilon
else:
new_bias_value = self.bias
else:
new_bias_value = None
if self.scalar_sigmas:
if self.optimize_sigmas:
if self.bias is not None:
print('Noisy convolution: sigma_conv={:2.4f}, sigma_bias={:2.4f}'.format(self.weight_sigma.item(),
self.bias_sigma.item()))
else:
print('Noisy convolution: sigma_conv={:2.4f}'.format(self.weight_sigma.item()))
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(effective_iter+1)*self.weight_sigma * weight_epsilon
else:
new_weight_value = self.weight
return F.conv1d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, self.dilation, self.groups)
else:
if self.training:
delta_weight = weight_epsilon
sz = self.weight_sigma.size()
for i in range(sz[0]):
delta_weight[i, ...] *= self.weight_sigma[i]
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(effective_iter+1)*delta_weight
else:
new_weight_value = self.weight
return F.conv1d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, self.dilation, self.groups)
[docs]class NoisyConv2d(_NoisyConvNd):
r"""Applies a 2D noisy convolution over an input signal composed of several input
planes.
In the simplest case, the output value of the layer with input size
:math:`(N, C_{in}, H, W)` and output :math:`(N, C_{out}, H_{out}, W_{out})`
can be precisely described as:
.. math::
\begin{equation*}
\text{out}(N_i, C_{out_j}) = \text{bias}(C_{out_j}) +
\sum_{k = 0}^{C_{in} - 1} \text{weight}(C_{out_j}, k) \star \text{input}(N_i, k)
\end{equation*},
where :math:`\star` is the valid 2D `cross-correlation`_ operator,
:math:`N` is a batch size, :math:`C` denotes a number of channels,
:math:`H` is a height of input planes in pixels, and :math:`W` is
width in pixels.
* :attr:`stride` controls the stride for the cross-correlation, a single
number or a tuple.
* :attr:`padding` controls the amount of implicit zero-paddings on both
sides for :attr:`padding` number of points for each dimension.
* :attr:`dilation` controls the spacing between the kernel points; also
known as the à trous algorithm. It is harder to describe, but this `link`_
has a nice visualization of what :attr:`dilation` does.
* :attr:`groups` controls the connections between inputs and outputs.
:attr:`in_channels` and :attr:`out_channels` must both be divisible by
:attr:`groups`. For example,
* At groups=1, all inputs are convolved to all outputs.
* At groups=2, the operation becomes equivalent to having two conv
layers side by side, each seeing half the input channels,
and producing half the output channels, and both subsequently
concatenated.
* At groups= :attr:`in_channels`, each input channel is convolved with
its own set of filters (of size
:math:`\left\lfloor\frac{\text{out_channels}}{\text{in_channels}}\right\rfloor`).
The parameters :attr:`kernel_size`, :attr:`stride`, :attr:`padding`, :attr:`dilation` can either be:
- a single ``int`` -- in which case the same value is used for the height and width dimension
- a ``tuple`` of two ints -- in which case, the first `int` is used for the height dimension,
and the second `int` for the width dimension
.. note::
Depending of the size of your kernel, several (of the last)
columns of the input might be lost, because it is a valid `cross-correlation`_,
and not a full `cross-correlation`_.
It is up to the user to add proper padding.
.. note::
The configuration when `groups == in_channels` and `out_channels == K * in_channels`
where `K` is a positive integer is termed in literature as depthwise convolution.
In other words, for an input of size :math:`(N, C_{in}, H_{in}, W_{in})`, if you want a
depthwise convolution with a depthwise multiplier `K`,
then you use the constructor arguments
:math:`(\text{in_channels}=C_{in}, \text{out_channels}=C_{in} * K, ..., \text{groups}=C_{in})`
Args:
in_channels (int): Number of channels in the input image
out_channels (int): Number of channels produced by the convolution
kernel_size (int or tuple): Size of the convolving kernel
stride (int or tuple, optional): Stride of the convolution. Default: 1
padding (int or tuple, optional): Zero-padding added to both sides of the input. Default: 0
dilation (int or tuple, optional): Spacing between kernel elements. Default: 1
groups (int, optional): Number of blocked connections from input channels to output channels. Default: 1
bias (bool, optional): If ``True``, adds a learnable bias to the output. Default: ``True``
Shape:
- Input: :math:`(N, C_{in}, H_{in}, W_{in})`
- Output: :math:`(N, C_{out}, H_{out}, W_{out})` where
.. math::
H_{out} = \left\lfloor\frac{H_{in} + 2 \times \text{padding}[0] - \text{dilation}[0]
\times (\text{kernel_size}[0] - 1) - 1}{\text{stride}[0]} + 1\right\rfloor
W_{out} = \left\lfloor\frac{W_{in} + 2 \times \text{padding}[1] - \text{dilation}[1]
\times (\text{kernel_size}[1] - 1) - 1}{\text{stride}[1]} + 1\right\rfloor
Attributes:
weight (Tensor): the learnable weights of the module of shape
(out_channels, in_channels, kernel_size[0], kernel_size[1])
bias (Tensor): the learnable bias of the module of shape (out_channels)
Examples::
>>> # With square kernels and equal stride
>>> m = nn.NoisyConv2d(16, 33, 3, stride=2)
>>> # non-square kernels and unequal stride and with padding
>>> m = nn.NoisyConv2d(16, 33, (3, 5), stride=(2, 1), padding=(4, 2))
>>> # non-square kernels and unequal stride and with padding and dilation
>>> m = nn.NoisyConv2d(16, 33, (3, 5), stride=(2, 1), padding=(4, 2), dilation=(3, 1))
>>> input = torch.randn(20, 16, 50, 100)
>>> output = m(input)
.. _cross-correlation:
https://en.wikipedia.org/wiki/Cross-correlation
.. _link:
https://github.com/vdumoulin/conv_arithmetic/blob/master/README.md
"""
def __init__(self, in_channels, out_channels, kernel_size, stride=1,
padding=0, dilation=1, groups=1, bias=True,
scalar_sigmas=True, optimize_sigmas=False, std_init=None, start_reducing_from_iter=25):
kernel_size = _pair(kernel_size)
stride = _pair(stride)
padding = _pair(padding)
dilation = _pair(dilation)
super(NoisyConv2d, self).__init__(
in_channels, out_channels, kernel_size, stride, padding, dilation,
False, _pair(0), groups, bias, scalar_sigmas, optimize_sigmas, std_init, start_reducing_from_iter)
[docs] def forward(self, input, iter=0):
weight_epsilon = MyTensor(*(self.out_channels, self.in_channels, *self.kernel_size)).normal_()
bias_epsilon = MyTensor(self.out_channels).normal_()
if self.bias is not None:
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_bias_value = self.bias + 1./(effective_iter+1)*self.bias_sigma * bias_epsilon
else:
new_bias_value = self.bias
else:
new_bias_value = None
if self.scalar_sigmas:
if self.optimize_sigmas:
if self.bias is not None:
print('Noisy convolution: sigma_conv={:2.4f}, sigma_bias={:2.4f}'.format(self.weight_sigma.item(),self.bias_sigma.item()))
else:
print('Noisy convolution: sigma_conv={:2.4f}'.format(self.weight_sigma.item()))
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(effective_iter+1)*self.weight_sigma*weight_epsilon
else:
new_weight_value = self.weight
return F.conv2d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, self.dilation, self.groups)
else:
if self.training:
delta_weight = weight_epsilon
sz = self.weight_sigma.size()
for i in range(sz[0]):
for j in range(sz[1]):
delta_weight[i,j,...] *= self.weight_sigma[i,j]
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(effective_iter+1)*delta_weight
else:
new_weight_value = self.weight
return F.conv2d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, self.dilation, self.groups)
[docs]class NoisyConv3d(_NoisyConvNd):
r"""Applies a 3D noisy convolution over an input signal composed of several input
planes.
In the simplest case, the output value of the layer with input size :math:`(N, C_{in}, D, H, W)`
and output :math:`(N, C_{out}, D_{out}, H_{out}, W_{out})` can be precisely described as:
.. math::
\begin{equation*}
\text{out}(N_i, C_{out_j}) = \text{bias}(C_{out_j}) +
\sum_{k = 0}^{C_{in} - 1} \text{weight}(C_{out_j}, k) \star \text{input}(N_i, k)
\end{equation*},
where :math:`\star` is the valid 3D `cross-correlation`_ operator
* :attr:`stride` controls the stride for the cross-correlation.
* :attr:`padding` controls the amount of implicit zero-paddings on both
sides for :attr:`padding` number of points for each dimension.
* :attr:`dilation` controls the spacing between the kernel points; also known as the à trous algorithm.
It is harder to describe, but this `link`_ has a nice visualization of what :attr:`dilation` does.
* :attr:`groups` controls the connections between inputs and outputs.
:attr:`in_channels` and :attr:`out_channels` must both be divisible by
:attr:`groups`. For example,
* At groups=1, all inputs are convolved to all outputs.
* At groups=2, the operation becomes equivalent to having two conv
layers side by side, each seeing half the input channels,
and producing half the output channels, and both subsequently
concatenated.
* At groups= :attr:`in_channels`, each input channel is convolved with
its own set of filters (of size
:math:`\left\lfloor\frac{\text{out_channels}}{\text{in_channels}}\right\rfloor`).
The parameters :attr:`kernel_size`, :attr:`stride`, :attr:`padding`, :attr:`dilation` can either be:
- a single ``int`` -- in which case the same value is used for the depth, height and width dimension
- a ``tuple`` of three ints -- in which case, the first `int` is used for the depth dimension,
the second `int` for the height dimension and the third `int` for the width dimension
.. note::
Depending of the size of your kernel, several (of the last)
columns of the input might be lost, because it is a valid `cross-correlation`_,
and not a full `cross-correlation`_.
It is up to the user to add proper padding.
.. note::
The configuration when `groups == in_channels` and `out_channels == K * in_channels`
where `K` is a positive integer is termed in literature as depthwise convolution.
In other words, for an input of size :math:`(N, C_{in}, D_{in}, H_{in}, W_{in})`, if you want a
depthwise convolution with a depthwise multiplier `K`,
then you use the constructor arguments
:math:`(\text{in_channels}=C_{in}, \text{out_channels}=C_{in} * K, ..., \text{groups}=C_{in})`
Args:
in_channels (int): Number of channels in the input image
out_channels (int): Number of channels produced by the convolution
kernel_size (int or tuple): Size of the convolving kernel
stride (int or tuple, optional): Stride of the convolution. Default: 1
padding (int or tuple, optional): Zero-padding added to all three sides of the input. Default: 0
dilation (int or tuple, optional): Spacing between kernel elements. Default: 1
groups (int, optional): Number of blocked connections from input channels to output channels. Default: 1
bias (bool, optional): If ``True``, adds a learnable bias to the output. Default: ``True``
Shape:
- Input: :math:`(N, C_{in}, D_{in}, H_{in}, W_{in})`
- Output: :math:`(N, C_{out}, D_{out}, H_{out}, W_{out})` where
.. math::
D_{out} = \left\lfloor\frac{D_{in} + 2 \times \text{padding}[0] - \text{dilation}[0]
\times (\text{kernel_size}[0] - 1) - 1}{\text{stride}[0]} + 1\right\rfloor
H_{out} = \left\lfloor\frac{H_{in} + 2 \times \text{padding}[1] - \text{dilation}[1]
\times (\text{kernel_size}[1] - 1) - 1}{\text{stride}[1]} + 1\right\rfloor
W_{out} = \left\lfloor\frac{W_{in} + 2 \times \text{padding}[2] - \text{dilation}[2]
\times (\text{kernel_size}[2] - 1) - 1}{\text{stride}[2]} + 1\right\rfloor
Attributes:
weight (Tensor): the learnable weights of the module of shape
(out_channels, in_channels, kernel_size[0], kernel_size[1], kernel_size[2])
bias (Tensor): the learnable bias of the module of shape (out_channels)
Examples::
>>> # With square kernels and equal stride
>>> m = nn.NoisyConv3d(16, 33, 3, stride=2)
>>> # non-square kernels and unequal stride and with padding
>>> m = nn.NoisyConv3d(16, 33, (3, 5, 2), stride=(2, 1, 1), padding=(4, 2, 0))
>>> input = torch.randn(20, 16, 10, 50, 100)
>>> output = m(input)
.. _cross-correlation:
https://en.wikipedia.org/wiki/Cross-correlation
.. _link:
https://github.com/vdumoulin/conv_arithmetic/blob/master/README.md
"""
def __init__(self, in_channels, out_channels, kernel_size, stride=1,
padding=0, dilation=1, groups=1, bias=True,
scalar_sigmas=True, optimize_sigmas=False, std_init=None, start_reducing_from_iter=25):
kernel_size = _triple(kernel_size)
stride = _triple(stride)
padding = _triple(padding)
dilation = _triple(dilation)
super(NoisyConv3d, self).__init__(
in_channels, out_channels, kernel_size, stride, padding, dilation,
False, _triple(0), groups, bias, scalar_sigmas, optimize_sigmas, std_init, start_reducing_from_iter)
[docs] def forward(self, input):
weight_epsilon = MyTensor(*(self.out_channels, self.in_channels, *self.kernel_size)).normal_()
bias_epsilon = MyTensor(self.out_channels).normal_()
if self.bias is not None:
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_bias_value = self.bias + 1./(effective_iter+1)*self.bias_sigma * bias_epsilon
else:
new_bias_value = self.bias
else:
new_bias_value = None
if self.scalar_sigmas:
if self.optimize_sigmas:
if self.bias is not None:
print('Noisy convolution: sigma_conv={:2.4f}, sigma_bias={:2.4f}'.format(self.weight_sigma.item(),
self.bias_sigma.item()))
else:
print('Noisy convolution: sigma_conv={:2.4f}'.format(self.weight_sigma.item()))
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(effective_iter+1)*self.weight_sigma * weight_epsilon
else:
new_weight_value = self.weight
return F.conv3d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, self.dilation, self.groups)
else:
if self.training:
delta_weight = weight_epsilon
sz = self.weight_sigma.size()
for i in range(sz[0]):
for j in range(sz[1]):
for k in range(sz[2]):
delta_weight[i, j, k, ...] *= self.weight_sigma[i, j, k]
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(effective_iter+1)*delta_weight
else:
new_weight_value = self.weight
return F.conv3d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, self.dilation, self.groups)
class _NoisyConvTransposeMixin(object):
def forward(self, input, output_size=None, iter=0):
output_padding = self._output_padding(input, output_size)
func = self._backend.ConvNd(
self.stride, self.padding, self.dilation, self.transposed,
output_padding, self.groups)
if self.bias is None:
return func(input, self.weight)
else:
return func(input, self.weight, self.bias)
def _output_padding(self, input, output_size):
if output_size is None:
return self.output_padding
output_size = list(output_size)
k = input.dim() - 2
if len(output_size) == k + 2:
output_size = output_size[-2:]
if len(output_size) != k:
raise ValueError(
"output_size must have {} or {} elements (got {})"
.format(k, k + 2, len(output_size)))
def dim_size(d):
return ((input.size(d + 2) - 1) * self.stride[d] -
2 * self.padding[d] + self.kernel_size[d])
min_sizes = [dim_size(d) for d in range(k)]
max_sizes = [min_sizes[d] + self.stride[d] - 1 for d in range(k)]
for size, min_size, max_size in zip(output_size, min_sizes, max_sizes):
if size < min_size or size > max_size:
raise ValueError((
"requested an output size of {}, but valid sizes range "
"from {} to {} (for an input of {})").format(
output_size, min_sizes, max_sizes, input.size()[2:]))
return tuple([output_size[d] - min_sizes[d] for d in range(k)])
[docs]class NoisyConvTranspose1d(_NoisyConvTransposeMixin, _NoisyConvNd):
r"""Applies a 1D noisy transposed convolution operator over an input image
composed of several input planes.
This module can be seen as the gradient of Conv1d with respect to its input.
It is also known as a fractionally-strided convolution or
a deconvolution (although it is not an actual deconvolution operation).
* :attr:`stride` controls the stride for the cross-correlation.
* :attr:`padding` controls the amount of implicit zero-paddings on both
sides for ``kernel_size - 1 - padding`` number of points. See note
below for details.
* :attr:`output_padding` controls the additional size added to one side
of the output shape. See note below for details.
* :attr:`dilation` controls the spacing between the kernel points; also known as the à trous algorithm.
It is harder to describe, but this `link`_ has a nice visualization of what :attr:`dilation` does.
* :attr:`groups` controls the connections between inputs and outputs.
:attr:`in_channels` and :attr:`out_channels` must both be divisible by
:attr:`groups`. For example,
* At groups=1, all inputs are convolved to all outputs.
* At groups=2, the operation becomes equivalent to having two conv
layers side by side, each seeing half the input channels,
and producing half the output channels, and both subsequently
concatenated.
* At groups= :attr:`in_channels`, each input channel is convolved with
its own set of filters (of size
:math:`\left\lfloor\frac{\text{out_channels}}{\text{in_channels}}\right\rfloor`).
.. note::
Depending of the size of your kernel, several (of the last)
columns of the input might be lost, because it is a valid `cross-correlation`_,
and not a full `cross-correlation`_.
It is up to the user to add proper padding.
.. note::
The :attr:`padding` argument effectively adds ``kernel_size - 1 - padding``
amount of zero padding to both sizes of the input. This is set so that
when a :class:`~torch.nn.Conv1d` and a :class:`~torch.nn.ConvTranspose1d`
are initialized with same parameters, they are inverses of each other in
regard to the input and output shapes. However, when ``stride > 1``,
:class:`~torch.nn.Conv1d` maps multiple input shapes to the same output
shape. :attr:`output_padding` is provided to resolve this ambiguity by
effectively increasing the calculated output shape on one side. Note
that :attr:`output_padding` is only used to find output shape, but does
not actually add zero-padding to output.
Args:
in_channels (int): Number of channels in the input image
out_channels (int): Number of channels produced by the convolution
kernel_size (int or tuple): Size of the convolving kernel
stride (int or tuple, optional): Stride of the convolution. Default: 1
padding (int or tuple, optional): ``kernel_size - 1 - padding`` zero-padding
will be added to both sides of the input. Default: 0
output_padding (int or tuple, optional): Additional size added to one side
of the output shape. Default: 0
groups (int, optional): Number of blocked connections from input channels to output channels. Default: 1
bias (bool, optional): If ``True``, adds a learnable bias to the output. Default: ``True``
dilation (int or tuple, optional): Spacing between kernel elements. Default: 1
Shape:
- Input: :math:`(N, C_{in}, L_{in})`
- Output: :math:`(N, C_{out}, L_{out})` where
.. math::
L_{out} = (L_{in} - 1) \times \text{stride} - 2 \times \text{padding}
+ \text{kernel_size} + \text{output_padding}
Attributes:
weight (Tensor): the learnable weights of the module of shape
(in_channels, out_channels, kernel_size[0], kernel_size[1])
bias (Tensor): the learnable bias of the module of shape (out_channels)
"""
def __init__(self, in_channels, out_channels, kernel_size, stride=1,
padding=0, output_padding=0, groups=1, bias=True, dilation=1,
scalar_sigmas=True, optimize_sigmas=False, std_init=None, start_reducing_from_iter=25):
kernel_size = _single(kernel_size)
stride = _single(stride)
padding = _single(padding)
dilation = _single(dilation)
output_padding = _single(output_padding)
super(NoisyConvTranspose1d, self).__init__(
in_channels, out_channels, kernel_size, stride, padding, dilation,
True, output_padding, groups, bias,
scalar_sigmas=scalar_sigmas, optimize_sigmas=optimize_sigmas, std_init=std_init,
start_reducing_from_iter=start_reducing_from_iter)
[docs] def forward(self, input, output_size=None, iter=0):
output_padding = self._output_padding(input, output_size)
weight_epsilon = MyTensor(*(self.out_channels, self.in_channels, *self.kernel_size)).normal_()
bias_epsilon = MyTensor(self.out_channels).normal_()
if self.bias is not None:
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_bias_value = self.bias + 1./(effective_iter+1)*self.bias_sigma * bias_epsilon
else:
new_bias_value = self.bias
else:
new_bias_value = None
if self.scalar_sigmas:
if self.optimize_sigmas:
if self.bias is not None:
print('Noisy convolution: sigma_conv={:2.4f}, sigma_bias={:2.4f}'.format(self.weight_sigma.item(),
self.bias_sigma.item()))
else:
print('Noisy convolution: sigma_conv={:2.4f}'.format(self.weight_sigma.item()))
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(1+effective_iter)*self.weight_sigma * weight_epsilon
else:
new_weight_value = self.weight
return F.conv_transpose1d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, output_padding, self.groups, self.dilation)
else:
if self.training:
delta_weight = weight_epsilon
sz = self.weight_sigma.size()
for i in range(sz[0]):
delta_weight[i, ...] *= self.weight_sigma[i]
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(1+effective_iter)*self.weight_sigma * weight_epsilon
else:
new_weight_value = self.weight
return F.conv_transpose1d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, output_padding, self.groups, self.dilation)
[docs]class NoisyConvTranspose2d(_NoisyConvTransposeMixin, _NoisyConvNd):
r"""Applies a 2D noisy transposed convolution operator over an input image
composed of several input planes.
This module can be seen as the gradient of Conv2d with respect to its input.
It is also known as a fractionally-strided convolution or
a deconvolution (although it is not an actual deconvolution operation).
* :attr:`stride` controls the stride for the cross-correlation.
* :attr:`padding` controls the amount of implicit zero-paddings on both
sides for ``kernel_size - 1 - padding`` number of points. See note
below for details.
* :attr:`output_padding` controls the additional size added to one side
of the output shape. See note below for details.
* :attr:`dilation` controls the spacing between the kernel points; also known as the à trous algorithm.
It is harder to describe, but this `link`_ has a nice visualization of what :attr:`dilation` does.
* :attr:`groups` controls the connections between inputs and outputs.
:attr:`in_channels` and :attr:`out_channels` must both be divisible by
:attr:`groups`. For example,
* At groups=1, all inputs are convolved to all outputs.
* At groups=2, the operation becomes equivalent to having two conv
layers side by side, each seeing half the input channels,
and producing half the output channels, and both subsequently
concatenated.
* At groups= :attr:`in_channels`, each input channel is convolved with
its own set of filters (of size
:math:`\left\lfloor\frac{\text{out_channels}}{\text{in_channels}}\right\rfloor`).
The parameters :attr:`kernel_size`, :attr:`stride`, :attr:`padding`, :attr:`output_padding`
can either be:
- a single ``int`` -- in which case the same value is used for the height and width dimensions
- a ``tuple`` of two ints -- in which case, the first `int` is used for the height dimension,
and the second `int` for the width dimension
.. note::
Depending of the size of your kernel, several (of the last)
columns of the input might be lost, because it is a valid `cross-correlation`_,
and not a full `cross-correlation`_.
It is up to the user to add proper padding.
.. note::
The :attr:`padding` argument effectively adds ``kernel_size - 1 - padding``
amount of zero padding to both sizes of the input. This is set so that
when a :class:`~torch.nn.Conv2d` and a :class:`~torch.nn.ConvTranspose2d`
are initialized with same parameters, they are inverses of each other in
regard to the input and output shapes. However, when ``stride > 1``,
:class:`~torch.nn.Conv2d` maps multiple input shapes to the same output
shape. :attr:`output_padding` is provided to resolve this ambiguity by
effectively increasing the calculated output shape on one side. Note
that :attr:`output_padding` is only used to find output shape, but does
not actually add zero-padding to output.
Args:
in_channels (int): Number of channels in the input image
out_channels (int): Number of channels produced by the convolution
kernel_size (int or tuple): Size of the convolving kernel
stride (int or tuple, optional): Stride of the convolution. Default: 1
padding (int or tuple, optional): ``kernel_size - 1 - padding`` zero-padding
will be added to both sides of each dimension in the input. Default: 0
output_padding (int or tuple, optional): Additional size added to one side
of each dimension in the output shape. Default: 0
groups (int, optional): Number of blocked connections from input channels to output channels. Default: 1
bias (bool, optional): If ``True``, adds a learnable bias to the output. Default: ``True``
dilation (int or tuple, optional): Spacing between kernel elements. Default: 1
Shape:
- Input: :math:`(N, C_{in}, H_{in}, W_{in})`
- Output: :math:`(N, C_{out}, H_{out}, W_{out})` where
.. math::
H_{out} = (H_{in} - 1) \times \text{stride}[0] - 2 \times \text{padding}[0]
+ \text{kernel_size}[0] + \text{output_padding}[0]
W_{out} = (W_{in} - 1) \times \text{stride}[1] - 2 \times \text{padding}[1]
+ \text{kernel_size}[1] + \text{output_padding}[1]
Attributes:
weight (Tensor): the learnable weights of the module of shape
(in_channels, out_channels, kernel_size[0], kernel_size[1])
bias (Tensor): the learnable bias of the module of shape (out_channels)
Examples::
>>> # With square kernels and equal stride
>>> m = nn.NoisyConvTranspose2d(16, 33, 3, stride=2)
>>> # non-square kernels and unequal stride and with padding
>>> m = nn.NoisyConvTranspose2d(16, 33, (3, 5), stride=(2, 1), padding=(4, 2))
>>> input = torch.randn(20, 16, 50, 100)
>>> output = m(input)
>>> # exact output size can be also specified as an argument
>>> input = torch.randn(1, 16, 12, 12)
>>> downsample = nn.NoisyConv2d(16, 16, 3, stride=2, padding=1)
>>> upsample = nn.NoisyConvTranspose2d(16, 16, 3, stride=2, padding=1)
>>> h = downsample(input)
>>> h.size()
torch.Size([1, 16, 6, 6])
>>> output = upsample(h, output_size=input.size())
>>> output.size()
torch.Size([1, 16, 12, 12])
.. _cross-correlation:
https://en.wikipedia.org/wiki/Cross-correlation
.. _link:
https://github.com/vdumoulin/conv_arithmetic/blob/master/README.md
"""
def __init__(self, in_channels, out_channels, kernel_size, stride=1,
padding=0, output_padding=0, groups=1, bias=True, dilation=1,
scalar_sigmas=True, optimize_sigmas=False, std_init=None, start_reducing_from_iter=25):
kernel_size = _pair(kernel_size)
stride = _pair(stride)
padding = _pair(padding)
dilation = _pair(dilation)
output_padding = _pair(output_padding)
super(NoisyConvTranspose2d, self).__init__(
in_channels, out_channels, kernel_size, stride, padding, dilation,
True, output_padding, groups, bias,
scalar_sigmas=scalar_sigmas, optimize_sigmas=optimize_sigmas, std_init=std_init,
start_reducing_from_iter=start_reducing_from_iter)
[docs] def forward(self, input, output_size=None, iter=0):
output_padding = self._output_padding(input, output_size)
weight_epsilon = MyTensor(*(self.out_channels, self.in_channels, *self.kernel_size)).normal_()
bias_epsilon = MyTensor(self.out_channels).normal_()
if self.bias is not None:
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_bias_value = self.bias + 1./(1+effective_iter)*self.bias_sigma * bias_epsilon
else:
new_bias_value = self.bias
else:
new_bias_value = None
if self.scalar_sigmas:
if self.optimize_sigmas:
if self.bias is not None:
print('Noisy convolution: sigma_conv={:2.4f}, sigma_bias={:2.4f}'.format(self.weight_sigma.item(),
self.bias_sigma.item()))
else:
print('Noisy convolution: sigma_conv={:2.4f}'.format(self.weight_sigma.item()))
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(1+effective_iter)*self.weight_sigma * weight_epsilon
else:
new_weight_value = self.weight
return F.conv_transpose2d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, output_padding, self.groups, self.dilation)
else:
if self.training:
delta_weight = weight_epsilon
sz = self.weight_sigma.size()
for i in range(sz[0]):
for j in range(sz[1]):
delta_weight[i, j, ...] *= self.weight_sigma[i, j]
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(1+effective_iter)*self.weight_sigma * weight_epsilon
else:
new_weight_value = self.weight
return F.conv_transpose2d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, output_padding, self.groups, self.dilation)
[docs]class NoisyConvTranspose3d(_NoisyConvTransposeMixin, _NoisyConvNd):
r"""Applies a 3D noisy transposed convolution operator over an input image composed of several input
planes.
The transposed convolution operator multiplies each input value element-wise by a learnable kernel,
and sums over the outputs from all input feature planes.
This module can be seen as the gradient of Conv3d with respect to its input.
It is also known as a fractionally-strided convolution or
a deconvolution (although it is not an actual deconvolution operation).
* :attr:`stride` controls the stride for the cross-correlation.
* :attr:`padding` controls the amount of implicit zero-paddings on both
sides for ``kernel_size - 1 - padding`` number of points. See note
below for details.
* :attr:`output_padding` controls the additional size added to one side
of the output shape. See note below for details.
* :attr:`dilation` controls the spacing between the kernel points; also known as the à trous algorithm.
It is harder to describe, but this `link`_ has a nice visualization of what :attr:`dilation` does.
* :attr:`groups` controls the connections between inputs and outputs.
:attr:`in_channels` and :attr:`out_channels` must both be divisible by
:attr:`groups`. For example,
* At groups=1, all inputs are convolved to all outputs.
* At groups=2, the operation becomes equivalent to having two conv
layers side by side, each seeing half the input channels,
and producing half the output channels, and both subsequently
concatenated.
* At groups= :attr:`in_channels`, each input channel is convolved with
its own set of filters (of size
:math:`\left\lfloor\frac{\text{out_channels}}{\text{in_channels}}\right\rfloor`).
The parameters :attr:`kernel_size`, :attr:`stride`, :attr:`padding`, :attr:`output_padding`
can either be:
- a single ``int`` -- in which case the same value is used for the depth, height and width dimensions
- a ``tuple`` of three ints -- in which case, the first `int` is used for the depth dimension,
the second `int` for the height dimension and the third `int` for the width dimension
.. note::
Depending of the size of your kernel, several (of the last)
columns of the input might be lost, because it is a valid `cross-correlation`_,
and not a full `cross-correlation`_.
It is up to the user to add proper padding.
.. note::
The :attr:`padding` argument effectively adds ``kernel_size - 1 - padding``
amount of zero padding to both sizes of the input. This is set so that
when a :class:`~torch.nn.Conv3d` and a :class:`~torch.nn.ConvTranspose3d`
are initialized with same parameters, they are inverses of each other in
regard to the input and output shapes. However, when ``stride > 1``,
:class:`~torch.nn.Conv3d` maps multiple input shapes to the same output
shape. :attr:`output_padding` is provided to resolve this ambiguity by
effectively increasing the calculated output shape on one side. Note
that :attr:`output_padding` is only used to find output shape, but does
not actually add zero-padding to output.
Args:
in_channels (int): Number of channels in the input image
out_channels (int): Number of channels produced by the convolution
kernel_size (int or tuple): Size of the convolving kernel
stride (int or tuple, optional): Stride of the convolution. Default: 1
padding (int or tuple, optional): ``kernel_size - 1 - padding`` zero-padding
will be added to both sides of each dimension in the input. Default: 0
output_padding (int or tuple, optional): Additional size added to one side
of each dimension in the output shape. Default: 0
groups (int, optional): Number of blocked connections from input channels to output channels. Default: 1
bias (bool, optional): If ``True``, adds a learnable bias to the output. Default: ``True``
dilation (int or tuple, optional): Spacing between kernel elements. Default: 1
Shape:
- Input: :math:`(N, C_{in}, D_{in}, H_{in}, W_{in})`
- Output: :math:`(N, C_{out}, D_{out}, H_{out}, W_{out})` where
.. math::
D_{out} = (D_{in} - 1) \times \text{stride}[0] - 2 \times \text{padding}[0]
+ \text{kernel_size}[0] + \text{output_padding}[0]
H_{out} = (H_{in} - 1) \times \text{stride}[1] - 2 \times \text{padding}[1]
+ \text{kernel_size}[1] + \text{output_padding}[1]
W_{out} = (W_{in} - 1) \times \text{stride}[2] - 2 \times \text{padding}[2]
+ \text{kernel_size}[2] + \text{output_padding}[2]
Attributes:
weight (Tensor): the learnable weights of the module of shape
(in_channels, out_channels, kernel_size[0], kernel_size[1], kernel_size[2])
bias (Tensor): the learnable bias of the module of shape (out_channels)
Examples::
>>> # With square kernels and equal stride
>>> m = nn.NoisyConvTranspose3d(16, 33, 3, stride=2)
>>> # non-square kernels and unequal stride and with padding
>>> m = nn.NoisyConv3d(16, 33, (3, 5, 2), stride=(2, 1, 1), padding=(0, 4, 2))
>>> input = torch.randn(20, 16, 10, 50, 100)
>>> output = m(input)
.. _cross-correlation:
https://en.wikipedia.org/wiki/Cross-correlation
.. _link:
https://github.com/vdumoulin/conv_arithmetic/blob/master/README.md
"""
def __init__(self, in_channels, out_channels, kernel_size, stride=1,
padding=0, output_padding=0, groups=1, bias=True, dilation=1,
scalar_sigmas=True, optimize_sigmas=False, std_init=None, start_reducing_from_iter=25):
kernel_size = _triple(kernel_size)
stride = _triple(stride)
padding = _triple(padding)
dilation = _triple(dilation)
output_padding = _triple(output_padding)
super(NoisyConvTranspose3d, self).__init__(
in_channels, out_channels, kernel_size, stride, padding, dilation,
True, output_padding, groups, bias,
scalar_sigmas=scalar_sigmas, optimize_sigmas=optimize_sigmas, std_init=std_init,
start_reducing_from_iter=start_reducing_from_iter)
[docs] def forward(self, input, output_size=None):
output_padding = self._output_padding(input, output_size)
weight_epsilon = MyTensor(*(self.out_channels, self.in_channels, *self.kernel_size)).normal_()
bias_epsilon = MyTensor(self.out_channels).normal_()
if self.bias is not None:
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_bias_value = self.bias + 1./(effective_iter+1)*self.bias_sigma * bias_epsilon
else:
new_bias_value = self.bias
else:
new_bias_value = None
if self.scalar_sigmas:
if self.optimize_sigmas:
if self.bias is not None:
print('Noisy convolution: sigma_conv={:2.4f}, sigma_bias={:2.4f}'.format(self.weight_sigma.item(),
self.bias_sigma.item()))
else:
print('Noisy convolution: sigma_conv={:2.4f}'.format(self.weight_sigma.item()))
if self.training:
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(effective_iter+1)*self.weight_sigma * weight_epsilon
else:
new_weight_value = self.weight
return F.conv_transpose3d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, output_padding, self.groups, self.dilation)
else:
if self.training:
delta_weight = weight_epsilon
sz = self.weight_sigma.size()
for i in range(sz[0]):
for j in range(sz[1]):
for k in range(sz[2]):
delta_weight[i, j, k, ...] *= self.weight_sigma[i, j, k]
effective_iter = max(0, iter - self.start_reducing_from_iter)
new_weight_value = self.weight + 1./(effective_iter+1)*self.weight_sigma * weight_epsilon
else:
new_weight_value = self.weight
return F.conv_transpose3d(input,
new_weight_value,
new_bias_value,
self.stride,
self.padding, output_padding, self.groups, self.dilation)