v0.1.0

First public release of TotalVariationImageFiltering.jl.

Added

• Core N-D TV problem API via TVProblem with configurable:
  • lambda and physical spacing
  • L2Fidelity and PoissonFidelity
  • IsotropicTV and AnisotropicTV
  • Neumann boundary handling
  • primal constraints (NoConstraint, NonnegativeConstraint, BoxConstraint)
• ROF solver (ROFConfig) for L2 + TV denoising using a Chambolle-style dual projected-gradient method.
• PDHG / Chambolle-Pock solver (PDHGConfig) for:
  • L2 + TV
  • Poisson KL + TV
  • optional primal constraints (non-negative and box-constrained reconstruction)
• Single-item and in-place solve APIs:
  • solve, solve!
  • reusable solver workspaces via ROFState and PDHGState
• Batch APIs for arrays shaped (spatial..., batch):
  • solve_batch, solve_batch!
  • summary and optional per-item solver statistics
• Automatic regularization parameter selection for ROF:
  • discrepancy principle (select_lambda_discrepancy)
  • Monte Carlo SURE (select_lambda_sure)
• Optional CUDA acceleration via extension (TotalVariationImageFilteringCUDAExt) when CUDA.jl is available.

Documentation

• Full docs site with guides for:
  • installation and quick start
  • problem/API usage
  • ROF and PDHG solver details
  • lambda selection
  • batch and CUDA usage
  • API reference and benchmarking

Validation

• Test suite covering:
  • problem/types/operators
  • ROF and PDHG solvers
  • single and batch APIs
  • lambda selection
  • CUDA extension behavior (when CUDA is available)
• CI configured for Julia 1.11 and 1.12.

Compatibility

• Julia 1.11+.
• CUDA support is optional and loaded through package extension when CUDA.jl and a functional CUDA runtime are present.

Notes

• ROFConfig currently supports L2Fidelity, Neumann boundary, and NoConstraint.
• PDHGConfig supports L2Fidelity and PoissonFidelity with NoConstraint, NonnegativeConstraint, or BoxConstraint.