resize¶

dragon.vision.resize(
  inputs,
  sizes=None,
  scales=None,
  mode='linear',
  align_corners=False,
  data_format='NCHW',
  **kwargs
)[source]¶

Resize input via interpolating neighborhoods.

sizes or scales will be selected by data_format:

x, sizes = dragon.ones((1, 2, 3, 4)), (6, 6)
a = dragon.vision.resize(x, sizes, data_format='NCHW')  # Shape: (1, 2, 6, 6)
c = dragon.vision.resize(x, sizes, data_format='NHWC')  # Shape: (1, 6, 6, 4)

Use align_corners to determine the input coordinates in linear interpolating:

# align_corners = False
# Use half-pixel transformation
scale = float(in_size) / float(out_size)
in_coord = (out_coord + 0.5) * scale - 0.5

# align_corners = True
# Use align-corners transformation
scale = float(in_size - 1) / float(out_size - 1)
in_coord = out_coord * scale

Parameters:

• inputs (dragon.Tensor) – The input tensor.
• sizes (Union[int, Sequence[int], dragon.Tensor], optional) – The output dimensions.
• scales (Union[float, Sequence[float], dragon.Tensor], optional) – The scale along each input dimension.
• mode (str, optional, default='nearest') – 'nearest' or 'linear'.
• align_corners (bool, optional, default=False) – Whether to align corners in linear interpolating.
• data_format (str, optional, default='NCHW') – 'NCHW' or 'NHWC'.

Returns:

dragon.Tensor – The output tensor.