dcnum.segm.segm_torch.torch_preproc
===================================

.. py:module:: dcnum.segm.segm_torch.torch_preproc


Functions
---------

.. autoapisummary::

   dcnum.segm.segm_torch.torch_preproc.preprocess_images


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

.. py:function:: preprocess_images(images: numpy.ndarray, norm_mean: float | None, norm_std: float | None, image_shape: tuple[int, int] | None = None)

   Transform image data to something torch models expect

   The transformation includes:

   - normalization (division by 255, subtraction of mean, division by std)
   - cropping and padding of the input images to `image_shape`. For padding,
     the median of each *individual* image is used.
   - casting the input images to four dimensions
     (batch_size, 1, height, width) where the second axis is "channels"

   :param images: Input image array (batch_size, height_in, width_in). If this is a
                  2D image, it will be reshaped to a 3D image with a batch_size of 1.
   :param norm_mean: Mean value used for standard score data normalization, i.e.
                     `normalized = `(images / 255 - norm_mean) / norm_std`; Set
                     to None to disable normalization.
   :param norm_std: Standard deviation used for standard score data normalization;
                    Set to None to disable normalization (see above).
   :param image_shape: Image shape for which the model was created (height, width).
                       If the image shape does not match the input image shape, then
                       the input images are padded/cropped to fit the image shape of
                       the model.

   :returns: 3D array with preprocessed image data of shape
             (batch_size, 1, height, width)
   :rtype: image_proc