mindspore.ops.DepthToSpace

class mindspore.ops.DepthToSpace(block_size)[source]

Rearranges blocks of depth data into spatial dimensions.

This is the reverse operation of SpaceToDepth.

The depth of output tensor is \(input\_depth / (block\_size * block\_size)\).

The output tensor’s height dimension is \(height * block\_size\).

The output tensor’s weight dimension is \(weight * block\_size\).

The input tensor’s depth must be divisible by block_size * block_size. The data format is “NCHW”.

Parameters

block_size (int) – The block size used to divide depth data. It must be >= 2.

Inputs:

  • x (Tensor) - The target tensor. It must be a 4-D tensor with shape \((N, C_{in}, H_{in}, W_{in})\). The data type is Number.

Outputs:

Tensor of shape \((N, C_{in} / \text{block_size} ^ 2, H_{in} * \text{block_size}, W_{in} * \text{block_size})\).

Raises

  • TypeError – If block_size is not an int.

  • ValueError – If block_size is less than 2.

  • ValueError – If length of shape of x is not equal to 4.

Supported Platforms:

Ascend GPU CPU

Examples

>>> x = Tensor(np.random.rand(1, 12, 1, 1), mindspore.float32)
>>> block_size = 2
>>> depth_to_space = ops.DepthToSpace(block_size)
>>> output = depth_to_space(x)
>>> print(output.shape)
(1, 3, 2, 2)