def __call__(self, inputs: Array) -> Array:
"""Applies a transposed convolution to the inputs. Behaviour mirrors of
`jax.lax.conv_transpose`.
Args:
inputs: input data with dimensions (batch, spatial_dims..., features).
Returns:
The convolved data.
"""
inputs = jnp.asarray(inputs, self.dtype)
strides = self.strides or (1,) * (inputs.ndim - 2)
in_features = inputs.shape[-1]
kernel_shape = self.kernel_size + (in_features, self.features)
kernel = self.param('kernel', self.kernel_init, kernel_shape)
kernel = jnp.asarray(kernel, self.dtype)
y = lax.conv_transpose(inputs,
kernel,
strides,
self.padding,
rhs_dilation=self.kernel_dilation,
precision=self.precision)
if self.use_bias:
bias = self.param('bias', self.bias_init, (self.features,))
bias = jnp.asarray(bias, self.dtype)
y = y + bias
return y
def apply_fun(params, inputs, **kwargs):
W = params
return lax.conv_transpose(inputs,
W,
strides,
padding,
dimension_numbers=dimension_numbers)
def __call__(self, inputs: jnp.ndarray) -> jnp.ndarray:
"""Computes the transposed convolution of the input.
Args:
inputs: An array of shape ``[spatial_dims, C]`` and rank-N+1 if unbatched,
or an array of shape ``[N, spatial_dims, C]`` and rank-N+2 if batched.
Returns:
An array of shape ``[spatial_dims, output_channels]`` and rank-N+1 if
unbatched, or an array of shape ``[N, spatial_dims, output_channels]``
and rank-N+2 if batched.
"""
unbatched_rank = self.num_spatial_dims + 1
allowed_ranks = [unbatched_rank, unbatched_rank + 1]
if inputs.ndim not in allowed_ranks:
raise ValueError(
f"Input to ConvNDTranspose needs to have rank in "
f"{allowed_ranks}, but input has shape {inputs.shape}.")
unbatched = inputs.ndim == unbatched_rank
if unbatched:
inputs = jnp.expand_dims(inputs, axis=0)
input_channels = inputs.shape[self.channel_index]
w_shape = self.kernel_shape + (self.output_channels, input_channels)
if self.mask is not None and self.mask.shape != w_shape:
raise ValueError("Mask needs to have the same shape as weights. "
f"Shapes are: {self.mask.shape}, {w_shape}")
w_init = self.w_init
if w_init is None:
fan_in_shape = self.kernel_shape + (input_channels, )
stddev = 1. / np.sqrt(np.prod(fan_in_shape))
w_init = hk.initializers.TruncatedNormal(stddev=stddev)
w = hk.get_parameter("w", w_shape, inputs.dtype, init=w_init)
if self.mask is not None:
w = w * self.mask
out = lax.conv_transpose(inputs,
w,
strides=self.stride,
padding=self.padding,
dimension_numbers=self.dimension_numbers)
if self.with_bias:
if self.channel_index == -1:
bias_shape = (self.output_channels, )
else:
bias_shape = (
self.output_channels, ) + (1, ) * self.num_spatial_dims
b = hk.get_parameter("b", bias_shape, init=self.b_init)
b = jnp.broadcast_to(b, out.shape)
out = out + b
if unbatched:
out = jnp.squeeze(out, axis=0)
return out
def __call__(self, x: JaxArray) -> JaxArray:
"""Returns the results of applying the transposed convolution to input x."""
y = lax.conv_transpose(x, self.w.value, self.strides, self.padding,
rhs_dilation=self.dilations,
dimension_numbers=('NCHW', 'HWIO', 'NCHW'), transpose_kernel=True)
if self.b:
y += self.b.value
return y
def conv_transpose(scope,
inputs,
features,
kernel_size,
strides=None,
padding='SAME',
kernel_dilation=None,
bias=True,
dtype=jnp.float32,
precision=None,
kernel_init=default_kernel_init,
bias_init=initializers.zeros):
"""Applies a transposed convolution to the inputs. Behaviour mirrors that of
`jax.lax.conv_transpose`.
Args:
scope: functional scope.
inputs: input data with dimensions (batch, spatial_dims..., features).
features: number of convolution filters.
kernel_size: shape of the convolutional kernel.
strides: a sequence of `n` integers, representing the inter-window
strides.
padding: either the string `'SAME'`, the string `'VALID'`, or a sequence
of `n` `(low, high)` integer pairs that give the padding to apply before
and after each spatial dimension.
kernel_dilation: `None`, or a sequence of `n` integers, giving the
dilation factor to apply in each spatial dimension of the convolution
kernel. Convolution with kernel dilation is also known as 'atrous
convolution'.
bias: whether to add a bias to the output (default: True).
dtype: the dtype of the computation (default: float32).
precision: numerical precision of the computation see `jax.lax.Precision`
for details.
kernel_init: initializer for the convolutional kernel.
bias_init: initializer for the bias.
Returns:
The convolved data.
"""
inputs = jnp.asarray(inputs, dtype)
strides = strides or (1, ) * (inputs.ndim - 2)
in_features = inputs.shape[-1]
kernel_shape = kernel_size + (in_features, features)
kernel = scope.param('kernel', kernel_init, kernel_shape)
kernel = jnp.asarray(kernel, dtype)
y = lax.conv_transpose(inputs,
kernel,
strides,
padding,
rhs_dilation=kernel_dilation,
precision=precision)
if bias:
bias = scope.param('bias', bias_init, (features, ))
bias = jnp.asarray(bias, dtype)
y = y + bias
return y
def _call_batched(self, x):
params, info = self.params, self.info
result = lax.conv_transpose(x,
params.kernel,
info.strides,
info.padding,
dimension_numbers=DIMENSION_NUMBERS)
if info.use_bias:
result += params.bias
return result
def convolution_transpose_op(self, params, inputs, **kwargs):
output = lax.conv_transpose(
inputs, params[0], self.strides, self.padding, dimension_numbers=self.dn
)
if self.use_bias:
output = jnp.add(output, params[1])
if self.activation:
output = self.activation(output)
return output
def call(self, inputs: np.ndarray) -> np.ndarray:
"""
Computes the transposed convolution of the input.
Args:
inputs: A rank-N+2 array with shape ``[N, spatial_dims, C]``.
Returns:
A rank-N+2 array with shape ``[N, spatial_dims, output_channels]``.
"""
required_rank = self.num_spatial_dims + 2
if inputs.ndim != required_rank:
raise ValueError(
f"Input to ConvND needs to have rank {required_rank}, "
f"but input has shape {inputs.shape}."
)
input_channels = inputs.shape[self.channel_index]
w_shape = self.kernel_shape + (self.output_channels, input_channels)
if self.mask is not None and self.mask.shape != w_shape:
raise ValueError(
"Mask needs to have the same shape as weights. "
f"Shapes are: {self.mask.shape}, {w_shape}"
)
w_init = self.w_init
if w_init is None:
fan_in_shape = self.kernel_shape + (input_channels,)
stddev = 1.0 / np.sqrt(np.prod(fan_in_shape))
w_init = initializers.TruncatedNormal(stddev=stddev)
w = hooks.get_parameter("w", w_shape, inputs.dtype, initializer=w_init)
if self.mask is not None:
w = w * self.mask
out = lax.conv_transpose(
inputs,
w,
strides=self.stride,
padding=self.padding,
dimension_numbers=self.dimension_numbers,
)
if self.with_bias:
if self.channel_index == -1:
bias_shape = (self.output_channels,)
else:
bias_shape = (self.output_channels,) + (1,) * self.num_spatial_dims
b = hooks.get_parameter("b", bias_shape, initializer=self.b_init)
b = jnp.broadcast_to(b, out.shape)
out = out + b
return out
def forward(self, x):
w, b = self.weights
x_shape = list(x.shape)
if len(x_shape) > 4:
self._check_nhwc()
new_batch_dim = functools.reduce(operator.mul, x.shape[:-3])
x = jnp.reshape(x, [new_batch_dim] + list(x.shape[-3:]))
res = lax.conv_transpose(x, w, self._strides, self._padding,
self._rhs_dilation,
self._dimension_numbers) + b
if len(x_shape) > 4:
res = jnp.reshape(res, x_shape[:-3] + list(res.shape[-3:]))
return res
def conv_transpose(inputs):
filter_shape_iter = iter(filter_shape)
kernel_shape = [out_chan if c == 'O' else
inputs.shape[lhs_spec.index('C')] if c == 'I' else
next(filter_shape_iter) for c in rhs_spec]
bias_shape = tuple(
itertools.dropwhile(lambda x: x == 1, [out_chan if c == 'C' else 1 for c in out_spec]))
kernel = parameter(kernel_shape, kernel_init, 'kernel')
bias = parameter(bias_shape, bias_init, 'bias')
return lax.conv_transpose(inputs, kernel, strides, padding,
dimension_numbers=dimension_numbers) + bias
def _upsample_nearest_neighbour(inputs_nchw):
# nearest neighbour upsampling on NCHW input
_n, input_c, h, w = inputs_nchw.shape
flat_inputs_shape = (-1, h, w, 1)
flat_inputs = jnp.reshape(inputs_nchw, flat_inputs_shape)
resize_kernel = jnp.ones((2, 2, 1, 1))
strides = (2, 2)
flat_outputs = conv_transpose(flat_inputs,
resize_kernel,
strides,
padding="SAME")
outputs_nchw_shape = (-1, input_c, 2 * h, 2 * w)
outputs_nchw = jnp.reshape(flat_outputs, outputs_nchw_shape)
return outputs_nchw
def __call__(self, inputs: Array) -> Array:
"""Applies a transposed convolution to the inputs. Behaviour mirrors of
`jax.lax.conv_transpose`.
Args:
inputs: input data with dimensions (batch, spatial_dims..., features).
Returns:
The convolved data.
"""
dtype = nkjax.maybe_promote_to_complex(self.dtype, inputs.dtype)
inputs = jnp.asarray(inputs, dtype)
if isinstance(self.kernel_size, int):
kernel_size = (self.kernel_size, )
else:
kernel_size = self.kernel_size
is_single_input = False
if inputs.ndim == len(kernel_size) + 1:
is_single_input = True
inputs = jnp.expand_dims(inputs, axis=0)
strides = self.strides or (1, ) * (inputs.ndim - 2)
in_features = inputs.shape[-1]
kernel_shape = kernel_size + (in_features, self.features)
kernel = self.param("kernel", self.kernel_init, kernel_shape,
self.dtype)
kernel = jnp.asarray(kernel, dtype)
y = lax.conv_transpose(
inputs,
kernel,
strides,
self.padding,
rhs_dilation=self.kernel_dilation,
precision=self.precision,
)
if is_single_input:
y = jnp.squeeze(y, axis=0)
if self.use_bias:
bias = self.param("bias", self.bias_init, (self.features, ),
self.dtype)
bias = jnp.asarray(bias, dtype)
bias = jnp.asarray(bias, dtype)
y = y + bias
return y
def __call__(self, inputs):
"""Connects Conv2DTranspose layer.
Args:
inputs: A rank-N+2 array with shape [N, spatial_dims, C].
Returns:
A rank-N+2 array with shape [N, spatial_dims, output_channels].
"""
if len(inputs.shape) != self._num_spatial_dims + 2:
raise ValueError(
"Input to ConvND needs to have rank {}, but input "
"has shape {}.".format(self._num_spatial_dims + 2,
inputs.shape))
weight_shape = self._kernel_shape + (inputs.shape[self._channel_index],
self._output_channels)
fan_in_shape = np.sqrt(np.prod(weight_shape[:-1]))
stddev = 1. / fan_in_shape
w_init = self._w_init or initializers.TruncatedNormal(stddev=stddev)
w = base.get_parameter("w", weight_shape, inputs.dtype, init=w_init)
if self._mask is not None:
if self._mask.shape != w.shape:
raise ValueError(
"Mask needs to have the same shape as weights. "
"Shapes are: {}, {}".format(self._mask.shape, w.shape))
w *= self._mask
result = lax.conv_transpose(inputs,
w,
self._stride,
self._padding,
dimension_numbers=self._dn)
if self._with_bias:
if self._channel_index == -1:
bias_shape = (self._output_channels, )
else:
bias_shape = (
self._output_channels, ) + (1, ) * self._num_spatial_dims
b = base.get_parameter("b", bias_shape, init=self._b_init)
result = result + b
return result
def batch_convolve_transpose(
input,
filter,
strides=1,
padding="VALID",
input_format=None,
filter_format=None,
output_format=None,
input_dilation=None,
filter_dilation=None,
transpose_kernel=False,
):
"""General n-dimensional convolution operator, with optional dilation.
Wraps Jax's conv_general_dilated functin, and thus also the XLA's `Conv
<https://www.tensorflow.org/xla/operation_semantics#conv_convolution>`_
operator.
Args:
input (Tensor): a rank `n+2` dimensional input array.
filter (Tensor): a rank `n+2` dimensional array of kernel weights.
strides (int, sequence of int, optional): a (sequence) of `n` integers,
representing the inter-window strides. If a scalar is given, it is
used `n` times. Defaults to `1`.
padding (sequence of couple, `'SAME'`, `'VALID'`, optional): a sequence of
`n` `(low, high)` integer pairs that give the padding to apply
before and after each spatial dimension. For `'VALID'`, those are
`0`. For `'SAME'`, they are the `input length - filter length + 1`
for each dim. Defaults to `'Valid'`.
input_format (`None` or str, optional): a string of same length as the
number of dimensions in `input` which specify their role
(see below). Defaults to `'NCW'` for 1d conv, `'NCHW'` for 2d conv,
and `'NDCHW'` for 3d conv.
input_dilation (`None`, int or sequence of int, optional): giving the
dilation factor to apply in each spatial dimension of `input`.
Inumpy.t dilation is also known as transposed convolution as it allows
to increase the output spatial dimension by inserting in the input
any number of `0`s between each spatial value.
filter_dilation (`None`, int or sequence of int): giving the dilation
factor to apply in each spatial dimension of `filter`. Filter
dilation is also known as atrous convolution as it corresponds to
inserting any number of `0`s in between the filter values, similar
to performing the non-dilated filter convolution with a subsample
version of the input across the spatial dimensions.
Returns:
Tensor: An array containing the convolution result.
Format of `input`, `filter` and `output`:
For example, to indicate dimension numbers consistent with the `conv` function
with two spatial dimensions, one could use `('NCHW', 'OIHW', 'NCHW')`. As
another example, to indicate dimension numbers consistent with the TensorFlow
Conv2D operation, one could use `('NHWC', 'HWIO', 'NHWC')`. When using the
latter form of convolution dimension specification, window strides are
associated with spatial dimension character labels according to the order in
which the labels appear in the `rhs_spec` string, so that `window_strides[0]`
is matched with the dimension corresponding to the first character
appearing in rhs_spec that is not `'I'` or `'O'`.
:param filter_format:
:param output_format:
:param transpose_kernel:
"""
# setting up the strides
if numpy.isscalar(strides):
strides = (strides,) * (input.ndim - 2)
elif len(strides) != (input.ndim - 2):
msg = "given strides: {} should match the number".format(
strides
) + "of spatial dim. in input: {}".format(input.ndim - 2)
raise ValueError(msg)
# setting up the padding
if type(padding) != str:
strides = (strides,) * (input.ndim - 2)
if len(padding) != (input.ndim - 2):
msg = "given padding: {} should match the ".format(
padding
) + "number of spatial dim. in input: {}".format(input.ndim - 2)
raise ValueError(msg)
# setting up the filter_format
if filter_format is None:
if filter.ndim == 3:
filter_format = "OIW"
elif filter.ndim == 4:
filter_format = "OIHW"
elif filter.ndim == 5:
filter_format = "OIDHW"
else:
msg = "filter_format should be given for >5 dimensions."
raise ValueError(msg)
elif len(filter_format) != filter.ndim:
msg = "given filter_format: {} should".format(
len(filter_format)
) + "match the number of dimension in filter: {}".format(filter.ndim)
raise ValueError(msg)
# setting up the input format
if input_format is None:
if len(filter.shape) == 3:
input_format = "NCW"
elif len(filter.shape) == 4:
input_format = "NCHW"
elif len(filter.shape) == 5:
input_format = "NCDHW"
else:
msg = "input_format should be given for >5 dimensions."
raise ValueError(msg)
elif len(input_format) != input.ndim:
msg = "given input_format: {} should".format(
len(input_format)
) + "match the number of dimension in input: {}".format(input.ndim)
raise ValueError(msg)
# setting up the output format
if output_format is None:
if len(filter.shape) == 3:
output_format = "NCW"
elif len(filter.shape) == 4:
output_format = "NCHW"
elif len(filter.shape) == 5:
output_format = "NCDHW"
else:
msg = "output_format should be given for >5 dimensions."
raise ValueError(msg)
elif len(output_format) != input.ndim:
msg = "given output_format: {} should".format(
len(output_format)
) + "match the number of dimension in output: {}".format(input.ndim)
raise ValueError(msg)
# setting up dilations
if numpy.isscalar(input_dilation):
input_dilation = (input_dilation,) * 2
if numpy.isscalar(filter_dilation):
filter_dilation = (filter_dilation,) * 2
specs = (input_format, filter_format, output_format)
return jla.conv_transpose(
lhs=input,
rhs=filter,
strides=strides,
padding=padding,
rhs_dilation=filter_dilation,
dimension_numbers=specs,
precision=None,
transpose_kernel=transpose_kernel,
)
def onnx_conv_transpose(x,
w,
b=None,
auto_pad='NOTSET',
dilations=None,
group=1,
kernel_shape=None,
output_padding=None,
output_shape=None,
pads=None,
strides=None,
**kwargs):
kernel_shape = kernel_shape or w.shape
spatial_size = w.ndim - 2
strides = strides or [1] * spatial_size
rhs_dilation = dilations or [1] * (w.ndim - 2)
# pad
if auto_pad == "NOTSET":
if pads is None:
pad_mode = 'VALID'
elif pads == 'VALID':
pad_mode = 'VALID'
elif pads == [0, 0] * spatial_size:
pad_mode = pads
else:
pad_mode = []
pad_pairs = len(pads) // 2
for idx in range(pad_pairs):
pad_mode.append((pads[idx], pads[idx + pad_pairs]))
elif auto_pad == "SAME_UPPER":
pad_mode = "SAME"
elif auto_pad == "VALID":
pad_mode = "VALID"
elif auto_pad == "SAME_LOWER":
raise NotImplemented("Conv with auto_pad `SAME_LOWER`")
else:
raise ValueError("Invalid auto_pad attribute: {}".format(auto_pad))
if b is not None:
b = b.reshape([1, w.shape[0]] + [1] * spatial_size)
else:
b = 0
res = lax.conv_transpose(
lhs=x,
rhs=w,
strides=strides,
padding=pad_mode,
rhs_dilation=rhs_dilation,
dimension_numbers=('NCHW', 'OIHW', 'NCHW'),
transpose_kernel=True,
precision=None,
)
# change output_padding order
# TODO
output_padding = ([0, 0, 0, 0] if output_padding is None else
[0, 0, output_padding[0], output_padding[1]])
if output_shape is not None:
need_append_output_pad = True
for spatial_idx in range(spatial_size):
total_pad = (output_padding[spatial_idx] +
output_padding[spatial_size + spatial_idx])
shape_diff = (output_shape[spatial_idx] -
res.shape[spatial_idx + 2] - total_pad)
if shape_diff == 0:
need_append_output_pad = False
else:
need_append_output_pad = True
if need_append_output_pad:
for spatial_idx in range(spatial_size):
shape_diff = output_shape[spatial_idx] - res.shape[spatial_idx
+ 2]
if shape_diff < 0:
raise Exception(
'output_sahpe can not samller than lax.conv_transpose output shape'
)
else:
output_padding[spatial_idx + spatial_size] += shape_diff
if output_padding != [0, 0, 0, 0]:
res = pad_helper(res, output_padding)
return [res + b]
