o yi@sddlmZmZddlZddlmZddlmZddlmZddl m Z ddl m Z m Z ddlmZerJdd lmZdd d Ze rIe esId dgZn GdddeZd dgZGdddeZdeded efddZGdd d e ZdS))AnyListN)Tensor)Module)Literal)Metric)_SKIP_SLOW_DOCTEST_try_proceed_with_timeout)_LPIPS_AVAILABLE)LPIPSreturncCstddddS)NTvgg) pretrainednet)_LPIPSrrK/home/ubuntu/.local/lib/python3.10/site-packages/torchmetrics/image/lpip.py_download_lpipssr%LearnedPerceptualImagePatchSimilarityr c@s eZdZdS)rN)__name__ __module__ __qualname__rrrrr#srcs&eZdZdeddffdd ZZS) NoTrainLpipsmoder cs tdS)zHthe network should not be able to be switched away from evaluation mode.F)supertrain)selfr __class__rrr*s zNoTrainLpips.train)rrrboolr __classcell__rrrrr)srimg normalizecCsD|r|dko |dkn|dk}|jdko!|jddko!|S)z1check that input is a valid image to the network.g?)maxminndimshape)r!r" value_checkrrr _valid_img/s r-c seZdZUdZdZeed<dZeed<dZeed<e e ed<e e ed<d gZ dd e d e ddededdf fdd Zde de ddfddZde fddZZS)ra The Learned Perceptual Image Patch Similarity (`LPIPS_`) is used to judge the perceptual similarity between two images. LPIPS essentially computes the similarity between the activations of two image patches for some pre-defined network. This measure has been shown to match human perception well. A low LPIPS score means that image patches are perceptual similar. Both input image patches are expected to have shape ``(N, 3, H, W)``. The minimum size of `H, W` depends on the chosen backbone (see `net_type` arg). .. note:: using this metrics requires you to have ``lpips`` package installed. Either install as ``pip install torchmetrics[image]`` or ``pip install lpips`` .. note:: this metric is not scriptable when using ``torch<1.8``. Please update your pytorch installation if this is a issue. As input to ``forward`` and ``update`` the metric accepts the following input - ``img1`` (:class:`~torch.Tensor`): tensor with images of shape ``(N, 3, H, W)`` - ``img2`` (:class:`~torch.Tensor`): tensor with images of shape ``(N, 3, H, W)`` As output of `forward` and `compute` the metric returns the following output - ``lpips`` (:class:`~torch.Tensor`): returns float scalar tensor with average LPIPS value over samples Args: net_type: str indicating backbone network type to use. Choose between `'alex'`, `'vgg'` or `'squeeze'` reduction: str indicating how to reduce over the batch dimension. Choose between `'sum'` or `'mean'`. normalize: by default this is ``False`` meaning that the input is expected to be in the [-1,1] range. If set to ``True`` will instead expect input to be in the ``[0,1]`` range. kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info. Raises: ModuleNotFoundError: If ``lpips`` package is not installed ValueError: If ``net_type`` is not one of ``"vgg"``, ``"alex"`` or ``"squeeze"`` ValueError: If ``reduction`` is not one of ``"mean"`` or ``"sum"`` Example: >>> import torch >>> _ = torch.manual_seed(123) >>> from torchmetrics.image.lpip import LearnedPerceptualImagePatchSimilarity >>> lpips = LearnedPerceptualImagePatchSimilarity(net_type='vgg') >>> # LPIPS needs the images to be in the [-1, 1] range. >>> img1 = (torch.rand(10, 3, 100, 100) * 2) - 1 >>> img2 = (torch.rand(10, 3, 100, 100) * 2) - 1 >>> lpips(img1, img2) tensor(0.3493, grad_fn=) Tis_differentiableFhigher_is_betterfull_state_update real_features fake_featuresralexmeannet_type reduction)sumr4r"kwargsr Nc stjdi|tstdd}||vr td|d|dt|dd|_d}||vr7td |d|||_t|t sFtd |||_ |j d t d d d|j dt d d ddS)NzzLPIPS metric requires that lpips is installed. Either install as `pip install torchmetrics[image]` or `pip install lpips`.)r r3squeezez#Argument `net_type` must be one of z , but got .F)rverbose)r4r7z$Argument `reduction` must be one of z/Argument `normalize` should be an bool but got sum_scoresr#r7)dist_reduce_fxtotalr)r__init__r ModuleNotFoundError ValueErrorrrr6 isinstancerr" add_statetorchtensor)rr5r6r"r8valid_net_typevalid_reductionrrrr?rs$ z.LearnedPerceptualImagePatchSimilarity.__init__img1img2c Cst||jr t||js7td|jd|jd||gd||gd|jr/ddgnddgd |j|||jd }|j| 7_|j |jd7_ d S) z(Update internal states with lpips score.zeExpected both input arguments to be normalized tensors with shape [N, 3, H, W]. Got input with shape z and z and values in range z+ when all values are expected to be in the rr&r$z range.)r"N) r-r"rAr+r)r(rr9r<r7r>)rrHrIlossrrrupdates"z,LearnedPerceptualImagePatchSimilarity.updatecCs*|jdkr |j|jS|jdkr|jSdS)z+Compute final perceptual similarity metric.r4r7N)r6r<r>)rrrrcomputes   z-LearnedPerceptualImagePatchSimilarity.compute)r3r4F)rrr__doc__r.r__annotations__r/r0rr__jit_ignored_attributes__strrrr?rKrLr rrrrr5s0  2      )r N)typingrrrDrtorch.nnrtyping_extensionsrtorchmetrics.metricrtorchmetrics.utilities.checksrr torchmetrics.utilities.importsr lpipsr rr__doctest_skip__rrr-rrrrrs&