Upsample - version 9#

This page documents version 9 of operator Upsample. See Upsample for the latest version (since version 10).

  • Domain: ai.onnx

  • Since version: 9

Upsample the input tensor. Each dimension value of the output tensor is:

output_dimension = floor(input_dimension * scale).

Inputs

  • X (T): N-D tensor

  • scales (tensor(float)): The scale array along each dimension. It takes value greater than or equal to 1. The number of elements of ‘scales’ should be the same as the rank of input ‘X’.

Outputs

  • Y (T): N-D tensor after resizing

Attributes

  • mode (string): Two interpolation modes: nearest (default), and linear (including bilinear, trilinear, etc)

Type Constraints

  • T: Constrain input ‘X’ and output ‘Y’ to all tensor types. Allowed types: tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8).

Examples#

test_cc_upsample_linear

Node:
  Upsample(X, scales) -> (Y)
  Attributes:
    mode = "linear"

Inputs:
  X: shape=(1, 1, 2, 2), dtype=float32
    [[[[1., 2.],
       [3., 4.]]]]
  scales: shape=(4,), dtype=float32
    [1., 1., 2., 2.]

Outputs:
  Y: shape=(1, 1, 4, 4), dtype=float32
    [[[[1. , 1.5, 2. , 2. ],
       [2. , 2.5, 3. , 3. ],
       [3. , 3.5, 4. , 4. ],
       [3. , 3.5, 4. , 4. ]]]]

test_cc_upsample_nearest

Node:
  Upsample(X, scales) -> (Y)
  Attributes:
    mode = "nearest"

Inputs:
  X: shape=(1, 1, 2, 2), dtype=float32
    [[[[1., 2.],
       [3., 4.]]]]
  scales: shape=(4,), dtype=float32
    [1., 1., 2., 3.]

Outputs:
  Y: shape=(1, 1, 4, 6), dtype=float32
    [[[[1., 1., 1., 2., 2., 2.],
       [1., 1., 1., 2., 2., 2.],
       [3., 3., 3., 4., 4., 4.],
       [3., 3., 3., 4., 4., 4.]]]]

test_cc_upsample_nearest_1d

Node:
  Upsample(X, scales) -> (Y)
  Attributes:
    mode = "nearest"

Inputs:
  X: shape=(3,), dtype=float32
    [10., 20., 30.]
  scales: shape=(1,), dtype=float32
    [2.]

Outputs:
  Y: shape=(6,), dtype=float32
    [10., 10., 20., 20., 30., 30.]

test_cc_upsample_nearest_default_mode

Node:
  Upsample(X, scales) -> (Y)

Inputs:
  X: shape=(1, 1, 2, 2), dtype=float32
    [[[[1., 2.],
       [3., 4.]]]]
  scales: shape=(4,), dtype=float32
    [1., 1., 2., 2.]

Outputs:
  Y: shape=(1, 1, 4, 4), dtype=float32
    [[[[1., 1., 2., 2.],
       [1., 1., 2., 2.],
       [3., 3., 4., 4.],
       [3., 3., 4., 4.]]]]

Differences with previous version (7)#

SchemaDiff: Upsample (domain 'ai.onnx')

  • old version: 7

  • new version: 9

  • breaking: yes

Breaking reasons:

  • input ‘scales’ (added): at position 1; option=Single; type_str=’tensor(float)’

  • attribute ‘scales’ (removed): type=FLOATS; required=True

Inputs:

  • [BREAKING] added ‘scales’: at position 1; option=Single; type_str=’tensor(float)’

Attributes:

  • [BREAKING] removed ‘scales’: type=FLOATS; required=True