def ref_deconvolution(x, w, b, base_axis, pad, stride, dilation, group):
y = []
for xx in x.reshape((-1,) + x.shape[base_axis:]):
y += [refs.deconvolution_2d(xx, w, b, pad, stride,
dilation, group)[np.newaxis]]
y = np.vstack(y)
return y.reshape(x.shape[:base_axis] + y.shape[1:])
def ref_deconvolution_2d(x,
w,
b,
base_axis,
pad,
stride,
dilation,
group,
channel_last=False,
output_padding=(0, 0)):
if channel_last:
transpose_x = refs.ChannelLastToFirstTranspose(x.ndim, len(pad))
transpose_w = refs.ChannelLastToFirstTranspose(w.ndim, len(pad))
return transpose_x.inv(
ref_deconvolution_2d(transpose_x(x), transpose_w(w), b, base_axis,
pad, stride, dilation, group, False,
output_padding))
y = []
for xx in x.reshape((-1, ) + x.shape[base_axis:]):
y += [
refs.deconvolution_2d(xx,
w,
b,
pad,
stride,
dilation,
group,
output_padding=output_padding)[np.newaxis]
]
y = np.vstack(y)
return y.reshape(x.shape[:base_axis] + y.shape[1:])
def ref_deconvolution(x, w, b, base_axis, pad, stride, dilation, group):
y = []
for xx in x.reshape((-1,) + x.shape[base_axis:]):
y += [refs.deconvolution_2d(xx, w, b, pad, stride,
dilation, group)[np.newaxis]]
y = np.vstack(y)
return y.reshape(x.shape[:base_axis] + y.shape[1:])
def ref_depthwise_deconvolution_2d(x, w, b, base_axis, pad, stride, dilation,
divisor):
""" implement depthwise deconvolution by using normal deconvolution
with group = inmaps """
# insert second dimension to weights
w = np.expand_dims(w, axis=1)
y = []
for xx in x.reshape((-1, ) + x.shape[base_axis:]):
groups = xx.shape[0] // divisor
yy = refs.deconvolution_2d(xx, w, b, pad, stride, dilation, groups)
y.append(yy[np.newaxis])
y = np.vstack(y)
return y.reshape(x.shape[:base_axis] + y.shape[1:])