depthwise_conv2d¶

dragon.vm.tensorflow.nn.depthwise_conv2d(
  input,
  filters,
  strides=1,
  padding='VALID',
  data_format='NHWC',
  dilations=1,
  name=None
)[source]¶

Apply the 2d depthwise convolution. [Chollet, 2016].

• If data_format is 'NCHW', excepts input shape \((N, C_{\text{in}}, H, W)\), filters shape \((C_{\text{out}}, 1, H_{\text{f}}, W_{\text{f}})\), and output shape is \((N, C_{\text{out}}, H_{\text{out}}, W_{\text{out}})\).
• If data_format is 'NHWC', excepts input shape \((N, H, W, C_{\text{in}})\), filters shape \((C_{\text{out}}, H_{\text{f}}, W_{\text{f}}, 1)\), and output shape is \((N, H_{\text{out}}, W_{\text{out}}, C_{\text{out}})\).
• padding could be 'VALID', 'SAME' or explicit padding size.

x = tf.ones((1, 2, 2, 2))
filters = tf.ones((2, 1, 1, 1))
y = tf.nn.depthwise_conv2d(x, filters)
assert y.shape == (1, 2, 2, 2)

Parameters:

• input (dragon.Tensor) – The input tensor.
• filters (dragon.Tensor) – The filters tensor.
• strides (Union[int, Sequence[int]]) – The stride of convolution window.
• padding (Union[int, Sequence[int], str]) – The padding algorithm or size.
• data_format (str, optional, default='NHWC') – 'NCHW' or 'NHWC'.
• dilations (Union[int, Sequence[int]], optional, default=1) – The rate of dilated filters.
• name (str, optional) – The operation name.

Returns:

dragon.Tensor – The output tensor.