[dependabot]

Updates the requirements on deepinv to permit the latest version.

Release notes

Sourced from deepinv's releases.

Release v0.5.0

New Features

• Add :func:deepinv.utils.load_raw to load a camera raw file as a Pytorch tensor along with a metadata dictionary containing ISP hyper-parameters (#1361 by Raphael Achddou)
• Add physics deepinv.physics.UltrasoundPlaneWave for ultrasound plane-wave imaging, which can simulate and reconstruct raw RF data. (#1338 by Adrien Besson)
• Add pretrained SwinIR weights for 2x super-resolution (#1304 by Vicky De Ridder)
• Add deepinv.sampling.PiGDMDataFidelity, deepinv.sampling.MomentMatchingDataFidelity, deepinv.sampling.ALDDataFidelity, deepinv.sampling.ScoreSDEDataFidelity and deepinv.sampling.ILVRDataFidelity noisy data-fidelity terms for diffusion posterior sampling, with a new example demo_noisy_data_fidelity.py (#1279 by Samuel Hurault)
• Publish the docs in llms.txt format using the sphinx-llm https://github.com/NVIDIA/sphinx-llm extension (#1362 by Julian Tachella)
• Add distributed backward propagation and training for samples too large to fit on a single device (#1088 by Benoît Malézieux)
• Add :func:deepinv.physics.TomographyWithAstra.from_astra_geometry to build the operator directly from pre-created astra geometries (#1102 by Margaret Duff)
• Add downloadable pretrained weights to deepinv.models.FFDNet (#1357 by Vicky De Ridder)
• Add :func:deepinv.utils.plot to disable image rescaling with rescale_mode=None. (#1339 by Delphine Doutsas)
• Add deepinv.datasets.CalgarySliceDataset for raw MRI data from Calgary-Campinas dataset(#1374 by Andrew Wang)
• Add deepinv.models.DIRECTModel pretrained models for MRI (Joint IC-Net, Recurrent VarNet, vSHARP, XPDNet etc.) (#1374 by Andrew Wang)
• Add :func:deepinv.utils.load_ismrmrd_raw to load data with ismrmrd library (#1374 by Andrew Wang)
• Add deepinv.physics.NonCartesianMRI non-Cartesian MRI physics with mri-nufft (#1381 by Andrew Wang)
• Add deepinv.datasets.DeteCTDataset dataset for real CT sinograms (#1378 by Andrew Wang)
• Add deepinv.datasets.Set5HR, deepinv.datasets.BSD100HR, deepinv.datasets.McMaster and deepinv.datasets.Kodak24 datasets (#1382 by Vicky De Ridder)
• Add ptychography acquisition geometries deepinv.physics.PtychographyGeometry, deepinv.physics.FarFieldPtychographyGeometry and deepinv.physics.NearFieldPtychographyGeometry, allowing to convert physical scan positions to pixel shifts (#1379 by Shantanu Kodgirwar)
• Add deepinv.models.WienerDeconvolution for closed-form Wiener deconvolution (#1315 by Ashish Verma)
• Add deepinv.optim.RDP prior (#1322 by Thibaut Modrzyk)
• Add deepinv.optim.BSREM algorithm for emission tomography and new PET demos (#1322 by Thibaut Modrzyk)
• Add blind Richardson-Lucy algorithm deepinv.optim.BlindRL for blind deconvolution along with a demo (#988 by Thibaut Modrzyk)
• Add colorblind palette and marker cyclers to :func:deepinv.utils.plotting.config_matplotlib (#1420 by Thibaut Modrzyk)
• Add deepinv.optim.SmoothedTVPrior prior (#1326 by Kushagra Shukla)
• Add deepinv.models.MPRNet restoration model, with pretrained weights for real-world denoising and deblurring (#1450 by Vicky De Ridder)
• Add deepinv.sampling.AncestralSolver, deepinv.sampling.DDPMSolver and deepinv.sampling.DDIMSolver, generalizing the DDPM and DDIM samplers to the SDE framework of deepinv.sampling.DiffusionSDE and deepinv.sampling.PosteriorDiffusion (#1445 by Minh Hai Nguyen)
• Add mask_first option to deepinv.physics.SpaceVaryingBlur and :func:deepinv.physics.functional.product_convolution2d (#1347 by Julian Tachella)

Changed

• deepinv.physics.Ptychography takesgeometry (#1379 by Shantanu Kodgirwar)
• (Breaking) change :meth:deepinv.physics.LinearPhysics.adjointness_test to compute a (more interpretable) relative error with respect to the operator norm (#1328 by Julian Tachella)
• Remove dependency on timm for SwinIR and SCUNet (#1304 by Vicky De Ridder)
• Center the titles above each figure in :func:deepinv.utils.plot_ortho3D (#1322 by Thibaut Modrzyk)
• Make the implementation of deepinv.transform.Shift parallel with respect to the number of transforms (#1408 by Jérémy Scanvic)
• (Breaking) deepinv.sampling.DPS now uses deepinv.sampling.DDPMSolver by default, which matches the original implementation. The previous Euler solver is available with solver="euler", and solver also accepts "ddpm", "ddim" or a solver instance (#1445 by Minh Hai Nguyen)
• Deprecate max_iter argument in deepinv.models.DEAL in favor of outer_iter and inner_iter (#1443 by Paul Bernard)
• Fix deepinv.loss.TVLoss to support 3D volumes, which previously ignored the last spatial dimension and were wrongly normalized (#1446 by Vicky De Ridder)
• Revert the scaling of diffusion sampling iterates back to [-1, 1] range. Also ensure inputs to physics are scaled back to [0,1] (#1413 by Minh Hai Nguyen)
• Add deepinv.transform.RotateViaShear transform class and deprecate the functional :func:deepinv.transform.rotate_via_shear API (#1436 by Jonathan Varela)

Fixed

• Apply random transforms consistently to paired ground truths and measurements in deepinv.datasets.ImageFolder and deepinv.datasets.HDF5Dataset (#1054 by Rusheel Sharma)
• Fix :func:deepinv.physics.PtychographyLinearOperator.A_adjoint corrected to consider the conjugate of the probe (#1353 by Shantanu Kodgirwar)
• Fix description of channels in documentation of deepinv.datasets.NBUDataset and provide link for more information on the dataset (#1348 by Delphine Doutsas)
• Fix inversion in deepinv.transform.Homography transforms (#1395 by Jérémy Scanvic)
• Fix global optim step size increased by failed backtracking in deepinv.optim.FixedPoint (#1314 by Thibaut Modrzyk)
• Fix incorrect shapes (H, W, D) to match deepinv's convention (D, H, W) in deepinv.datasets.BrainWebPET and deepinv.physics.PET (#1322 by Thibaut Modrzyk)
• clamp_output=False would not work in deepinv.models.DEAL (#1443 by Paul Bernard)
• Fix the mask_first=False pretrained deepinv.models.KernelIdentificationNetwork (#1347 by Julian Tachella)
• Raise informative error when filter is not set in deepinv.physics.Blur and deepinv.physics.BlurFFT (#1337 by Romain Vo)

... (truncated)

Commits

• da772e3 Adds ptychography geometry for allowing experimental data (#1379)
• 98f0697 correct the changelog (#1468)
• dda2640 feat: add shear rotation transform (#1436)
• 96c2fe4 Add Wiener deconvolution (#1315)
• 812452d better adjointness test (#1328)
• 0b5466b Allow any maintainer adding new self-hosted GPU runners (#1426)
• 91dbd9c Add MPRNet (#1450)
• 7dc687a fix matplotlib figure leak (#1461)
• 6494773 Docs: Move MICCAI '26 and hackathon to past events (#1455)
• 0caba6f add a RAW file loader in the io that returns a torch Tensor along with an ISP...
• Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)

[dependabot]

Labels

The following labels could not be found: dependencies, python. Please create them before Dependabot can add them to a pull request.

Please fix the above issues or remove invalid values from dependabot.yml.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!