channel_shuffle¶

dragon.nn.channel_shuffle(
  inputs,
  axis=- 1,
  group=1,
  **kwargs
)[source]¶

Apply the group shuffle to each channel of input. [Zhang et.al, 2017].

Examples:

x = dragon.constant([1, 2, 3, 4])
print(dragon.nn.channel_shuffle(x, group=2))  # [1, 3, 2, 4]

Parameters:

• inputs (dragon.Tensor) – The inputs.
• axis (int, optional, default=-1) – The channel axis.
• group (int, optional, default=1) – The number of shuffle groups.

Returns:

dragon.Tensor – The output tensor.