pyiqa.archs.dmm_arch
====================

.. py:module:: pyiqa.archs.dmm_arch

.. autoapi-nested-parse::

   Debiased Mapping for Full-Reference Image Quality Assessment

   @article{chen2025debiased,
     title={Debiased mapping for full-reference image quality assessment},
     author={Chen, Baoliang and Zhu, Hanwei and Zhu, Lingyu and Wang, Shanshe and Pan, Jingshan and Wang, Shiqi},
     journal={IEEE Transactions on Multimedia},
     year={2025},
     publisher={IEEE}
   }

   Reference:
       - Arxiv link: https://ieeexplore.ieee.org/abstract/document/10886996
       - Official Github: https://github.com/Baoliang93/DMM







Module Contents
---------------

.. py:data:: names

.. py:class:: FeaturesExtractor(target_features=('relu3_3', 'relu4_3'), use_input_norm=False, requires_grad=False, replace_pooling=True)

   Bases: :py:obj:`torch.nn.Module`


   VGG16 feature extractor for DMM.

   :param target_features: VGG feature names to collect.
   :type target_features: tuple[str, ...]
   :param use_input_norm: Whether to apply ImageNet normalization.
   :type use_input_norm: bool
   :param requires_grad: Whether extracted backbone parameters are trainable.
   :type requires_grad: bool
   :param replace_pooling: Whether to replace max-pooling with L2 pooling.
   :type replace_pooling: bool


   .. py:method:: forward(x)

      Extract configured feature maps from input image tensor.



   .. py:method:: replace_pooling(module: torch.nn.Module) -> torch.nn.Module


.. py:class:: L2Pool2d(kernel_size: int = 3, stride: int = 2, padding=1)

   Bases: :py:obj:`torch.nn.Module`


   Applies L2 pooling with Hann window of size 3x3
   :param x: Tensor with shape (N, C, H, W)


   .. py:method:: forward(x: torch.Tensor) -> torch.Tensor

      Apply L2 pooling using a Hann filter kernel.



.. py:class:: DMM(reduce_dim=256, kernel_size=5, features_to_compute=('relu3_3', 'relu4_3'), criterion=torch.nn.CosineSimilarity(), use_dropout=True, **kwargs)

   Bases: :py:obj:`torch.nn.Module`


   DMM full-reference IQA model.

   :param reduce_dim: Reserved compatibility argument.
   :type reduce_dim: int
   :param kernel_size: Reserved compatibility argument.
   :type kernel_size: int
   :param features_to_compute: Feature names used for DMM.
   :type features_to_compute: tuple[str, ...]
   :param criterion: Reserved compatibility argument.
   :type criterion: torch.nn.Module
   :param use_dropout: Reserved compatibility argument.
   :type use_dropout: bool
   :param \*\*kwargs: Reserved compatibility arguments.


   .. py:method:: forward(Dist, Ref, as_loss=False)

      Compute DMM score or training loss.

      :param Dist: Distorted image tensor ``(N, 3, H, W)``.
      :type Dist: torch.Tensor
      :param Ref: Reference image tensor ``(N, 3, H, W)``.
      :type Ref: torch.Tensor
      :param as_loss: If ``True``, return mean score for optimization.
      :type as_loss: bool

      :returns: Per-sample scores when ``as_loss=False``, otherwise a
                scalar loss tensor.
      :rtype: torch.Tensor



   .. py:method:: prepare_image_adt(tensor_image)

      Adaptively resize large images for stable patch-based scoring.