|
|
|
import warnings |
|
|
|
import torch |
|
import torch.nn.functional as F |
|
|
|
|
|
def resize(input, |
|
size=None, |
|
scale_factor=None, |
|
mode='nearest', |
|
align_corners=None, |
|
warning=True): |
|
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`') |
|
if isinstance(size, torch.Size): |
|
size = tuple(int(x) for x in size) |
|
return F.interpolate(input, size, scale_factor, mode, align_corners) |
|
|