mindspore.ops.MaxPoolWithArgmaxV2

View Source On Gitee
class mindspore.ops.MaxPoolWithArgmaxV2(kernel_size, strides=None, pads=0, dilation=(1, 1), ceil_mode=False, argmax_type=mstype.int64)[source]

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

Typically the input is of shape \((N_{in}, C_{in}, H_{in}, W_{in})\), MaxPool outputs regional maximum in the \((H_{in}, W_{in})\)-dimension. Given kernel size \((h_{ker}, w_{ker})\) and stride \((s_0, s_1)\), the operation is as follows:

\[\text{output}(N_i, C_j, h, w) = \max_{m=0, \ldots, h_{ker}-1} \max_{n=0, \ldots, w_{ker}-1} \text{input}(N_i, C_j, s_0 \times h + m, s_1 \times w + n)\]

Warning

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

Parameters

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

  • strides (Union[int, tuple[int]], optional) – The distance of kernel moving, an int number that represents not only the height of movement but also the width of movement, or a tuple of two int numbers that represent height and width of movement respectively. Default: None , meaning that strides = kernel_size.

  • pads (Union[int, tuple[int]], optional) – An int number that represents the depth, height and width of movement are both strides, or a tuple of two int numbers that represent depth, height and width of movement respectively. Default: 0.

  • dilation (Union[int, tuple[int]], optional) – Control the stride of elements in the kernel. Default: (1, 1) .

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

  • argmax_type (mindspore.dtype, optional) – The dtype for argmax. Default: mstype.int64 . [Disabled in Ascend.]

Inputs:

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

Outputs:

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

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

    \[H_{out} = \left\lfloor\frac{H_{in} + 2 * \text{pads[0]} - \text{dilation[0]} \times (\text{kernel_size[0]} - 1) - 1}{\text{strides[0]}} + 1\right\rfloor\]
    \[W_{out} = \left\lfloor\frac{W_{in} + 2 * \text{pads[1]} - \text{dilation[1]} \times (\text{kernel_size[1]} - 1) - 1}{\text{strides[1]}} + 1\right\rfloor\]

  • argmax (Tensor) - Index corresponding to the maximum value. Data type is int32 or int64 in GPU and CPU, is uint16 in Ascend.

Raises

  • TypeError – If x is not a Tensor.

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

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

  • ValueError – If kernel_size, strides or dilation is less than 1.

  • ValueError – If pads is less than 0.

  • ValueError – If pads is more than half of kernel_size.

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

  • TypeError – If ceil_mode is not bool.

Supported Platforms:

Ascend GPU CPU

Examples

>>> import mindspore
>>> import numpy as np
>>> from mindspore import Tensor, ops
>>> x = Tensor(np.arange(20 * 16 * 50 * 32).reshape((20, 16, 50, 32)), mindspore.float32)
>>> maxpool_arg_v2_op = ops.MaxPoolWithArgmaxV2(kernel_size=(3, 2), strides=(2, 1))
>>> output_tensor, argmax = maxpool_arg_v2_op(x)
>>> print(output_tensor.shape)
(20, 16, 24, 31)
>>> print(argmax.shape)
(20, 16, 24, 31)