View source on GitHub |
Depthwise 2-D convolution.
tf.compat.v1.nn.depthwise_conv2d(
input,
filter,
strides,
padding,
rate=None,
name=None,
data_format=None,
dilations=None
)
Given a 4D input tensor ('NHWC' or 'NCHW' data formats)
and a filter tensor of shape
[filter_height, filter_width, in_channels, channel_multiplier]
containing in_channels
convolutional filters of depth 1, depthwise_conv2d
applies a different filter to each input channel (expanding from 1 channel
to channel_multiplier
channels for each), then concatenates the results
together. The output has in_channels * channel_multiplier
channels.
In detail, with the default NHWC format,
output[b, i, j, k * channel_multiplier + q] = sum_{di, dj}
filter[di, dj, k, q] * input[b, strides[1] * i + rate[0] * di,
strides[2] * j + rate[1] * dj, k]
Must have strides[0] = strides[3] = 1
. For the most common case of the
same horizontal and vertical strides, strides = [1, stride, stride, 1]
.
If any value in rate
is greater than 1, we perform atrous depthwise
convolution, in which case all values in the strides
tensor must be equal
to 1.
Usage Example:
x = np.array([
[1., 2.],
[3., 4.],
[5., 6.]
], dtype=np.float32).reshape((1, 3, 2, 1))
kernel = np.array([
[1., 2.],
[3., 4]
], dtype=np.float32).reshape((2, 1, 1, 2))
tf.compat.v1.nn.depthwise_conv2d(x, kernel, strides=[1, 1, 1, 1],
padding='VALID').numpy()
array([[[[10., 14.],
[14., 20.]],
[[18., 26.],
[22., 32.]]]], dtype=float32)
tf.compat.v1.nn.depthwise_conv2d(x, kernel, strides=[1, 1, 1, 1],
padding=[[0, 0], [1, 0], [1, 0], [0, 0]]
).numpy()
array([[[[ 0., 0.],
[ 3., 4.],
[ 6., 8.]],
[[ 0., 0.],
[10., 14.],
[14., 20.]],
[[ 0., 0.],
[18., 26.],
[22., 32.]]]], dtype=float32)
Returns | |
---|---|
A 4-D Tensor with shape according to data_format . E.g., for
"NHWC" format, shape is
[batch, out_height, out_width, in_channels * channel_multiplier].
|