mindspore.ops.batch_to_space_nd

View Source On Gitee
mindspore.ops.batch_to_space_nd(input_x, block_shape, crops)[source]

Divides batch dimension with blocks and interleaves these blocks back into spatial dimensions.

This operation will divide batch dimension N into blocks with block_shape, the output tensor’s N dimension is the corresponding number of blocks after division. The output tensor’s \(w_1, ..., w_M\) dimension is the product of original \(w_1, ..., w_M\) dimension and block_shape with given amount to crop from dimension, respectively.

If the input shape is \((n, c_1, ... c_k, w_1, ..., w_M)\), the output shape is \((n', c_1, ... c_k, w'_1, ..., w'_M)\), where

\[\begin{split}\begin{array}{ll} \\ n' = n//(block\_shape[0]*...*block\_shape[M-1]) \\ w'_i = w_i*block\_shape[i-1]-crops[i-1][0]-crops[i-1][1] \end{array}\end{split}\]
Parameters

  • input_x (Tensor) – The input tensor. It must be greater or equal to 2-D tensor(equal to 4-D tensor on Ascend), batch dimension must be divisible by product of block_shape.

  • block_shape (Union[list(int), tuple(int), int]) – The block shape of dividing block with all value greater than or equal to 1. If block_shape is a tuple or list, the length of block_shape is M corresponding to the number of spatial dimensions. If block_shape is an int, the block size of M dimensions are the same, equal to block_shape. In this case of Ascend, M must be 2.

  • crops (Union[list(int), tuple(int)]) – The crops values for spatial dimensions, containing M subtraction list. Each contains 2 integer values. All values must be >= 0. crops[i] specifies the crops values for spatial dimension i, which corresponds to input dimension i + offset,where offset = N-M, and N is the number of input dimensions. It is required that \(input\_shape[i+offset]*block\_shape[i] > crops[i][0]+crops[i][1]\)

Returns

Tensor, the output tensor with the same type as input.

Raises

  • TypeError – If block_shape is not one of list, tuple, int.

  • TypeError – If crops is neither list nor tuple.

  • ValueError – If block_shape is not one dimensional when block_shape is a list or tuple.

  • ValueError – If the length of block_shape is not 2 on Ascend.

  • ValueError – If the element of block_shape is not an integer larger than or euqal to 1.

  • ValueError – If shape of crops is not (M, 2), where M is the length of block_shape.

  • ValueError – If the element of crops is not an integer larger than or euqal to 0.

Supported Platforms:

Ascend GPU CPU

Examples

>>> import mindspore
>>> import numpy as np
>>> from mindspore import Tensor, ops
>>> block_shape = [2, 2]
>>> crops = [[0, 0], [0, 0]]
>>> input_x = Tensor(np.array([[[[1]]], [[[2]]], [[[3]]], [[[4]]]]), mindspore.float32)
>>> output = ops.batch_to_space_nd(input_x, block_shape, crops)
>>> print(output)
[[[[1.  2.]
   [3.  4.]]]]