INNER CODE UNIT · Python

lpips

nv-tlabs/3dgrut · thirdparty/lpipsPyTorch/__init__.py:21

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)

View source record →