mindspore.ops.MaxPool3DWithArgmax

class mindspore.ops.MaxPool3DWithArgmax(ksize, strides, pads, dilation=(1, 1, 1), ceil_mode=False, data_format='NCDHW', argmax_type=mstype.int64)

Performs a 3D max pooling on the input Tensor and returns both max values and indices.

Typically the input is a Tensor with shape $(N_{in},C_{in},D_{in},H_{in},W_{in})$, outputs regional maximum in the $(D_{in},H_{in},W_{in})$-dimension. Given ksize $ks=(d_{ker},h_{ker},w_{ker})$ and strides $s=(s_0,s_1,s_2)$, the operation is as follows.

$$\text{output}(N_i,C_j,d,h,w)=\underset{l=0,\dots ,d_{ker}-1}{max}\underset{m=0,\dots ,h_{ker}-1}{max}\underset{n=0,\dots ,w_{ker}-1}{max}\text{input}(N_i,C_j,s_0\times d+l,s_1\times h+m,s_2\times w+n)$$

The output is a Tensor with shape $(N_{out},C_{out},D_{out},H_{out},W_{out})$ and its depth, height and width are:

$$\begin{array}{r}\begin{array}{l}\\ D_{out}=\frac{D_{in}+2\times \text{pads}[0]-\text{dilation}[0]\times (\text{ksize}[0]-1)-1}{\text{stride}[0]}+1\\ H_{out}=\frac{H_{in}+2\times \text{pads}[1]-\text{dilation}[1]\times (\text{ksize}[1]-1)-1}{\text{stride}[1]}+1\\ W_{out}=\frac{W_{in}+2\times \text{pads}[2]-\text{dilation}[2]\times (\text{ksize}[2]-1)-1}{\text{stride}[2]}+1\end{array}\end{array}$$

Warning

This is an experimental API that is subject to change or deletion.

Parameters

• ksize (Union[int, tuple[int]]) – The size of kernel used to take the maximum value and arg value, is an int number that represents depth, height and width of the kernel, or a tuple of three int numbers that represent depth, height and width respectively.

• strides (Union[int, tuple[int]]) – The distance of kernel moving, an int number that represents the depth, height and width of movement are both strides, or a tuple of three int numbers that represent depth, height and width of movement respectively.

• pads (Union[int, tuple[int]]) – An int number that represents the depth, height and width of movement are both strides, or a tuple of three int numbers that represent depth, height and width of movement respectively.

• dilation (Union[int, tuple[int]]) – Default: (1, 1, 1) .

• ceil_mode (bool) – Whether to use ceil instead of floor to calculate output shape. Default: False .

• data_format (str) – The optional value for data format. Currently only support 'NCDHW' . Default: 'NCDHW' .

• argmax_type (mindspore.dtype) – The dtype for argmax. Default: mstype.int64 .

Inputs:

• x (Tensor) - Tensor of shape $(N_{in},C_{in},D_{in},H_{in},W_{in})$ with data type of int8, int16, int32, int64, uint8, uint16, uint32, uint64, float16, float32 or float64.

Outputs:

Tuple of 2 Tensors, representing the maxpool result and where the max values are generated.

• output (Tensor) - Maxpooling result, with shape $(N_{out},C_{out},D_{out},H_{out},W_{out})$. It has the same data type as x.

• argmax (Tensor) - Index corresponding to the maximum value. Data type is int32 or int64.

Raises

• TypeError – If x is not a Tensor.

• ValueError – If length of shape of x is not equal to 5.

• TypeError – If ksize , strides , pads or dilation is not int or tuple.

• ValueError – If ksize or strides is less than 1.

• ValueError – If pads is less than 0.

• ValueError – If data_format is not 'NCDHW'.

• ValueError – If argmax_type is not mindspore.int64 or mindspore.int32.

Supported Platforms:

Ascend GPU CPU

Examples

>>> import mindspore
>>> import numpy as np
>>> from mindspore import Tensor, ops
>>> x = Tensor(np.arange(2 * 1 * 2 * 2 * 2).reshape((2, 1, 2, 2, 2)), mindspore.float32)
>>> max_pool3d_with_arg_op = ops.MaxPool3DWithArgmax(ksize=2, strides=1, pads=1)
>>> output_tensor, argmax = max_pool3d_with_arg_op(x)
>>> print(output_tensor.shape)
(2, 1, 3, 3, 3)
>>> print(argmax.shape)
(2, 1, 3, 3, 3)