| 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) | |