|
|
|
|
|
|
|
|
|
|
|
import warnings |
|
|
|
import torch.nn.functional as F |
|
|
|
|
|
def resize(input, size=None, scale_factor=None, mode="nearest", align_corners=None, warning=False): |
|
if warning: |
|
if size is not None and align_corners: |
|
input_h, input_w = tuple(int(x) for x in input.shape[2:]) |
|
output_h, output_w = tuple(int(x) for x in size) |
|
if output_h > input_h or output_w > output_h: |
|
if ( |
|
(output_h > 1 and output_w > 1 and input_h > 1 and input_w > 1) |
|
and (output_h - 1) % (input_h - 1) |
|
and (output_w - 1) % (input_w - 1) |
|
): |
|
warnings.warn( |
|
f"When align_corners={align_corners}, " |
|
"the output would more aligned if " |
|
f"input size {(input_h, input_w)} is `x+1` and " |
|
f"out size {(output_h, output_w)} is `nx+1`" |
|
) |
|
return F.interpolate(input, size, scale_factor, mode, align_corners) |
|
|