TensorFlow MRI 0.23.1

Release 0.23.0

This release bumps the supported TF version to v2.10.x. No other new
features or bug fixes are included.

v0.23.0

Tutorials added

TensorFlow MRI 0.22.0

Release 0.22.0

This release bumps the supported TF version to v2.10.x. No other new
features or bug fixes are included.

TensorFlow MRI 0.21.0

Release 0.21.0

This release contains new functionality for wavelet decomposition and
reconstruction and optimized Gram matrices for some linear operators. It also
redesigns the convex optimization module and contains some improvements to the
documentation.

Breaking Changes

• tfmri.convex:

  • Argument ndim has been removed from all functions.
  • All functions will now require the domain dimension to be
    specified. Therefore, domain_dimension is now the first positional
    argument in several functions including ConvexFunctionIndicatorBall,
    ConvexFunctionNorm and ConvexFunctionTotalVariation. However, while
    this parameter is no longer optional, it is now possible to pass dynamic
    or static information as opposed to static only (at least in the general
    case, but specific operators may have additional restrictions).
  • For consistency and accuracy, argument axis of
    ConvexFunctionTotalVariation has been renamed to axes.

Major Features and Improvements

• tfmri.convex:

  • Added new class ConvexFunctionL1Wavelet, which enables image/signal
    reconstruction with L1-wavelet regularization.
  • Added new argument gram_operator to ConvexFunctionLeastSquares,
    which allows the user to specify a custom, potentially more efficient Gram
    matrix.

• tfmri.linalg:

  • Added new classes LinearOperatorNUFFT and LinearOperatorGramNUFFT
    to enable the use of NUFFT as a linear operator.
  • Added new class LinearOperatorWavelet to enable the use of wavelets
    as a linear operator.

• tfmri.sampling:

  • Added new ordering type sorted_half to radial_trajectory.

• tfmri.signal:

  • Added new functions wavedec and waverec for wavelet decomposition
    and reconstruction, as well as utilities wavelet_coeffs_to_tensor,
    tensor_to_wavelet_coeffs, and max_wavelet_level.

Bug Fixes and Other Changes

• tfmri.recon:

  • Improved error reporting for least_squares.

TensorFlow MRI 0.20.0

Release 0.20.0

Major Features and Improvements

• tfmri.layers:

  • Added new layers MaxPooling1D, MaxPooling2D, MaxPooling3D,
    AveragePooling1D, AveragePooling2D and AveragePooling3D.
    These are drop-in replacements for the core Keras layers of the same name,
    but they also support complex values.
  • Added new layers DWT1D, DWT2D, DWT3D, IDWT1D, IDWT2D,
    and IDWT3D to compute 1D, 2D and 3D forward and inverse discrete wavelet
    transforms.
  • Layer ConvBlock is now deprecated in favor of the new endpoints in
    the tfmri.models submodule.
  • Layer UNet is now deprecated in favor of the new endpoints in
    the tfmri.models submodule.

• tfmri.models:

  • Added new models ConvBlock1D, ConvBlock2D and ConvBlock3D. These
    replace the previous ConvBlock layer, which is now deprecated.
  • Added new models UNet1D, UNet2D and UNet3D. These replace
    the previous UNet layer, which is now deprecated.

Bug Fixes and Other Changes

• tfmri.signal:

  • Improved static shape inference for dwt op.

TensorFlow MRI 0.19.0

Release 0.19.0

This release bumps the supported TensorFlow version to 2.9.

Major Features and Improvements

• tfmri.image:

  • Added new arguments batch_dims and image_dims to
    image_gradients, gradient_magnitude, psnr, ssim and
    ssim_multiscale.
  • Argument rank of psnr, ssim and ssim_multiscale is now
    deprecated. To update, use image_dims instead.
  • image_gradients and gradient_magnitude now support complex inputs.

• tfmri.losses:

  • Added new arguments batch_dims and image_dims to all image quality
    losses. Argument rank is now deprecated (use image_dims instead).
  • Added new arguments multichannel and complex_part to all image
    quality losses. These mirror the behaviour of the corresponding metrics
    and allows for single-channel inputs and extraction of parts from complex
    numbers, respectively.
  • The canonical API names StructuralSimilarityLoss and
    MultiscaleStructuralSimilarityLoss have been
    changed to SSIMLoss and SSIMMultiscaleLoss, respectively. The
    old names are still available, but may be removed in the future.

• tfmri.metrics:

  • Image quality metrics can now accept complex inputs without also specifying
    complex_part, in which case the unmodified complex values will be passed
    to the downstream function. This may not be supported for all metrics.
  • Added new arguments batch_dims and image_dims to all image quality
    metrics. Argument rank is now deprecated (use image_dims instead).
  • The canonical API names PeakSignalToNoiseRatio,
    StructuralSimilarity and MultiscaleStructuralSimilarity have been
    changed to PSNR, SSIM and SSIMMultiscale, respectively. The
    old names are still available, but may be removed in the future.

• tfmri.recon:

  • Added new argument preserve_phase to tfmri.recon.pf. This allows
    the user to recover the phase as well as the magnitude during partial
    Fourier reconstruction. Argument return_complex has the same behaviour
    and is now deprecated.
  • Added new aliases adjoint (for adj), least_squares
    (for lstsq) and partial_fourier (for pf). These are now the
    canonical aliases, but the old ones will still be supported.

• tfmri.signal:

  • Added new ops dwt and idwt to compute the N-dimensional discrete
    wavelet transform and its inverse, respectively.

• tfmri.plot:

  • Added new argument norm to image_sequence, tiled_image and
    tiled_image_sequence. This allows the user to specify the scaling
    to be applied before the colormap.

Bug Fixes and Other Changes

• Fixed a bug with k-space weighting in homodyne detection method of
  tfmri.recon.partial_fourier.
• Like core TensorFlow, we now compile with _GLIBCXX_USE_CXX11_ABI=1.
• Like core TensorFlow, Python wheels now conform to manylinux2014, an
  upgrade from manylinux2010.

TensorFlow MRI 0.18.0

Release 0.18.0

Major Features and Improvements

• tfmri.convex:

  • All references to ndim in ConvexFunction and its subclasses have
    been deprecated in favor of domain_dimension or domain_shape. This
    applies to constructor arguments, properties and methods.
  • Refactored static and dynamic shape properties and methods to single-source
    shape information on the _shape and _shape_tensor methods.
    domain_dimension, domain_dimension_tensor, batch_shape and
    batch_shape_tensor now just call shape and shape_tensor and
    should not be overriden.

• tfmri.initializers:

  • New module with initializers VarianceScaling, GlorotNormal,
    GlorotUniform, HeNormal, HeUniform, LecunNormal and
    LecunUniform. All initializers are drop-in replacements for their
    Keras counterparts and support complex values.

• tfmri.io:

  • New function parse_twix to parse TWIX RAID files (Siemens raw data).

• tfmri.layers:

  • Added new layers Conv1D, Conv2D and Conv3D. All layers
    are drop-in replacements for their Keras counterparts and support
    complex values.

Bug Fixes and Other Changes

• Using a new API import system, which should address some issues with
  the autocomplete features of some editors.

TensorFlow MRI 0.17.0

Release 0.17.0

Breaking Changes

• tfmri.signal:

  • Renamed keyword argument traj of filter_kspace and crop_kspace
    to trajectory.

Major Features and Improvements

• tfmri.plot:

  • New utility tiled_image.

TensorFlow MRI 0.16.0

Release 0.16.0

Breaking Changes

• tfmri.convex:

  • Several of the inputs and outputs of admm_minimize have been renamed
    to improve clarity and to make the interface more consistent with the
    tfmri.optimize module.

Major Features and Improvements

• tfmri.convex:

  • admm_minimize now supports batches of inputs.

  • admm_minimize has two new arguments f_prox_kwargs and
    g_prox_kwargs that allow passing additional arguments to the proximal
    operators of f and g.

  • admm_minimize has a new argument name that allows specifying
    the name of the operation.

  • Method _prox of ConvexFunctionQuadratic has a new argument
    solver_kwargs that allows passing additional arguments to the
    internal solver.

  • New properties shape and batch_shape for ConvexFunction and
    its subclasses. These allow retrieval of static shape information.

  • New methods ndim_tensor, shape_tensor and batch_shape_tensor
    for ConvexFunction and its subclasses. These allow retrieval of the
    dynamic shape information.

• tfmri.linalg:

  • New argument bypass_gradient for conjugate_gradient.

• tfmri.recon:

  • lstsq has a new argument return_optimizer_state which allows the
    user to retrieve the internal optimizer state.

• tfmri.optimize:

  • New function gradient_descent implementing the gradient descent method
    for minimization of differentiable functions.

• tfmri.plot:

  • New argument dpi for functions image_sequence and
    tiled_image_sequence.
  • New function close, alias of matplotlib.pyplot.close.

Bug Fixes and Other Changes

• tfmri.convex:

  • Fixed a bug in method prox of ConvexFunctionNorm,
    ConvexFunctionL2Squared and ConvexFunctionQuadratic that caused
    errors when running in graph mode.

• tfmri.linalg:

  • Fixed a bug in internal linear algebra framework that would cause errors
    when running in graph mode.

• tfmri.plot:

  • Removed lazy import mechanism which was causing problems when using
    matplotlib outside TFMRI. For now we use regular imports.

TensorFlow MRI 0.15.0

Release 0.15.0

Major Features and Improvements

• tfmri.plot:

  • image_sequence and tiled_image_sequence now support different
    layout mechanisms via the argument layout.
  • image_sequence and tiled_image_sequence now support display
    bounding boxes via the arguments bbox_inches and pad_inches.