espnet2.enh.layers.ncsnpp_utils.up_or_down_sampling.naive_upsample_2d

source

espnet2.enh.layers.ncsnpp_utils.up_or_down_sampling.naive_upsample_2d(x, factor=2)

Perform naive 2D upsampling on a tensor.

This function takes an input tensor representing a batch of 2D images and upsamples each image by the specified factor using nearest-neighbor interpolation. The output tensor will have its height and width multiplied by the upsampling factor.

• Parameters:
  • x (torch.Tensor) – Input tensor of shape [N, C, H, W], where:
    • N: Number of images in the batch.
    • C: Number of channels in each image.
    • H: Height of each image.
    • W: Width of each image.
  • factor (int) – Upsampling factor. Default is 2.
• Returns: Output tensor of shape [N, C, H * factor, W * factor].
• Return type: torch.Tensor

Examples

>>> x = torch.rand(1, 3, 4, 4)  # A batch of 1 image with 3 channels
>>> upsampled = naive_upsample_2d(x, factor=2)
>>> print(upsampled.shape)
torch.Size([1, 3, 8, 8])  # The output shape after upsampling

NOTE

This method is a simple and efficient way to increase the resolution of images but may not preserve the quality as well as more advanced methods.