Spaces:
Running
on
L40S
Running
on
L40S
File size: 1,174 Bytes
31f2f28 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 |
# Copyright (c) OpenMMLab. All rights reserved.
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)
|