mindspore.nn.Pad

class mindspore.nn.Pad(paddings, mode='CONSTANT')[source]

Pads the input tensor according to the paddings and mode.

Parameters
  • paddings (tuple) –

    The shape of parameter paddings is (N, 2). N is the rank of input data. All elements of paddings are int type. For D th dimension of the x, paddings[D, 0] indicates how many sizes to be extended ahead of the D th dimension of the input tensor, and paddings[D, 1] indicates how many sizes to be extended behind of the D th dimension of the input tensor. The padded size of each dimension D of the output is: \(paddings[D, 0] + input\_x.dim\_size(D) + paddings[D, 1]\), e.g.:

    mode = "CONSTANT".
    paddings = [[1,1], [2,2]].
    x = [[1,2,3], [4,5,6], [7,8,9]].
    # The above can be seen: 1st dimension of `x` is 3, 2nd dimension of `x` is 3.
    # Substitute into the formula to get:
    # 1st dimension of output is paddings[0][0] + 3 + paddings[0][1] = 1 + 3 + 1 = 5.
    # 2nd dimension of output is paddings[1][0] + 3 + paddings[1][1] = 2 + 3 + 2 = 7.
    # So the shape of output is (5, 7).
    

  • mode (str) – Specifies padding mode. The optional values are “CONSTANT”, “REFLECT”, “SYMMETRIC”. Default: “CONSTANT”.

Inputs:
  • x (Tensor) - The input tensor.

Outputs:

Tensor, the tensor after padding.

  • If mode is “CONSTANT”, it fills the edge with 0, regardless of the values of the x. If the x is [[1,2,3], [4,5,6], [7,8,9]] and paddings is [[1,1], [2,2]], then the Outputs is [[0,0,0,0,0,0,0], [0,0,1,2,3,0,0], [0,0,4,5,6,0,0], [0,0,7,8,9,0,0], [0,0,0,0,0,0,0]].

  • If mode is “REFLECT”, it uses a way of symmetrical copying through the axis of symmetry to fill in. If the x is [[1,2,3], [4,5,6], [7,8,9]] and paddings is [[1,1], [2,2]], then the Outputs is [[6,5,4,5,6,5,4], [3,2,1,2,3,2,1], [6,5,4,5,6,5,4], [9,8,7,8,9,8,7], [6,5,4,5,6,5,4]].

  • If mode is “SYMMETRIC”, the filling method is similar to the “REFLECT”. It is also copied according to the symmetry axis, except that it includes the symmetry axis. If the x is [[1,2,3], [4,5,6], [7,8,9]] and paddings is [[1,1], [2,2]], then the Outputs is [[2,1,1,2,3,3,2], [2,1,1,2,3,3,2], [5,4,4,5,6,6,5], [8,7,7,8,9,9,8], [8,7,7,8,9,9,8]].

Raises
  • TypeError – If paddings is not a tuple.

  • ValueError – If length of paddings is more than 4 or its shape is not (n, 2).

  • ValueError – If mode is not one of ‘CONSTANT’, ‘REFLECT’, ‘SYMMETRIC’.

Supported Platforms:

Ascend GPU CPU

Examples

>>> from mindspore import Tensor
>>> import mindspore.nn as nn
>>> import numpy as np
>>> # If `mode` is "CONSTANT"
>>> class Net(nn.Cell):
...     def __init__(self):
...         super(Net, self).__init__()
...         self.pad = nn.Pad(paddings=((1, 1), (2, 2)), mode="CONSTANT")
...     def construct(self, x):
...         return self.pad(x)
>>> x = Tensor(np.array([[1, 2, 3], [4, 5, 6]]), mindspore.float32)
>>> pad = Net()
>>> output = pad(x)
>>> print(output)
[[0. 0. 0. 0. 0. 0. 0.]
 [0. 0. 1. 2. 3. 0. 0.]
 [0. 0. 4. 5. 6. 0. 0.]
 [0. 0. 0. 0. 0. 0. 0.]]
>>> # Another way to call
>>> pad = ops.Pad(paddings=((1, 1), (2, 2)))
>>> # From the above code, we can see following:
>>> # "paddings=((1, 1), (2, 2))",
>>> # paddings[0][0] = 1, indicates a row of values is filled top of the input data in the 1st dimension.
>>> # Shown as follows:
>>> # [[0. 0. 0.]
>>> #  [1. 2. 3.]
>>> #  [4. 5. 6.]]
>>> # paddings[0][1] = 1 indicates a row of values is filled below input data in the 1st dimension.
>>> # Shown as follows:
>>> # [[0. 0. 0.]
>>> #  [1. 2. 3.]
>>> #  [4. 5. 6.]
>>> #  [0. 0. 0.]]
>>> # paddings[1][0] = 2, indicates 2 rows of values is filled in front of input data in the 2nd dimension.
>>> # Shown as follows:
>>> # [[0. 0. 0. 0. 0.]
>>> #  [0. 0. 1. 2. 3.]
>>> #  [0. 0. 4. 5. 6.]
>>> #  [0. 0. 0. 0. 0.]]
>>> # paddings[1][1] = 2, indicates 2 rows of values is filled in front of input data in the 2nd dimension.
>>> # Shown as follows:
>>> # [[0. 0. 0. 0. 0. 0. 0.]
>>> #  [0. 0. 1. 2. 3. 0. 0.]
>>> #  [0. 0. 4. 5. 6. 0. 0.]
>>> #  [0. 0. 0. 0. 0. 0. 0.]]
>>> output = pad(x)
>>> print(output)
[[0. 0. 0. 0. 0. 0. 0.]
 [0. 0. 1. 2. 3. 0. 0.]
 [0. 0. 4. 5. 6. 0. 0.]
 [0. 0. 0. 0. 0. 0. 0.]]
>>> # if mode is "REFLECT"
>>> class Net(nn.Cell):
...     def __init__(self):
...         super(Net, self).__init__()
...         self.pad = nn.Pad(paddings=((1, 1), (2, 2)), mode="REFLECT")
...     def construct(self, x):
...         return self.pad(x)
>>> x = Tensor(np.array([[1, 2, 3], [4, 5, 6]]), mindspore.float32)
>>> pad = Net()
>>> output = pad(x)
>>> print(output)
[[6. 5. 4. 5. 6. 5. 4.]
 [3. 2. 1. 2. 3. 2. 1.]
 [6. 5. 4. 5. 6. 5. 4.]
 [3. 2. 1. 2. 3. 2. 1.]]
>>> # if mode is "SYMMETRIC"
>>> class Net(nn.Cell):
...     def __init__(self):
...         super(Net, self).__init__()
...         self.pad = nn.Pad(paddings=((1, 1), (2, 2)), mode="SYMMETRIC")
...     def construct(self, x):
...         return self.pad(x)
>>> x = Tensor(np.array([[1, 2, 3], [4, 5, 6]]), mindspore.float32)
>>> pad = Net()
>>> output = pad(x)
>>> print(output)
[[2. 1. 1. 2. 3. 3. 2.]
 [2. 1. 1. 2. 3. 3. 2.]
 [5. 4. 4. 5. 6. 6. 5.]
 [5. 4. 4. 5. 6. 6. 5.]]