| import torch | |
| from .modules.lpips import LPIPS | |
| def lpips(x: torch.Tensor, | |
| y: torch.Tensor, | |
| net_type: str = 'alex', | |
| version: str = '0.1'): | |
| r"""Function that measures | |
| Learned Perceptual Image Patch Similarity (LPIPS). | |
| Arguments: | |
| x, y (torch.Tensor): the input tensors to compare. | |
| net_type (str): the network type to compare the features: | |
| 'alex' | 'squeeze' | 'vgg'. Default: 'alex'. | |
| version (str): the version of LPIPS. Default: 0.1. | |
| """ | |
| device = x.device | |
| criterion = LPIPS(net_type, version).to(device) | |
| return criterion(x, y) | |