v1.0.7 release

What's Changed

• Added full vectorized support for batched inputs with tissue masking, eliminating ragged-array execution caused by masked regions.
• Introduced automatic structural, textural, and color-based hashing for cache keys when caching is enabled but explicit user-defined identifiers are not provided.

Full Changelog: v1.0.5...v1.0.7

V1.0.5 Release

(1) Accelerate dictionary learning by torch.compile. Remove the graph-breaking behaviors in certain computation procedures.
(2) Improved dictionary learning performance by caching rss gradient terms in ISTA/FISTA of code steps and pre-computation in residual calculations of dict steps.
(3) Refactoring. Cleaning of the method signatures.

Full Changelog: v1.0.4...v1.0.5

V1.0.4 Release

(1) Fix an issue such that the augmentation may draw out-of-intended-boundary noise for concentration augmentation.

(2) Add optional least square solver (torch.linalg.lstsq) for concentration computation (concentration_method='ls', besides the existing 'ista' and 'cd' options) when dealing with huge number of small image inputs (e.g., batches of Nx3x256x256 input) on-the-fly. Note that while 'ls' is faster in such scenario, it does not have sparse constraints, and it may fail on GPU if the height/width of image is too large regardless of batch size, due to the limitation of torch.linalg.lstsq.

(3) Readme and demo update.

V1.0.3 Release

• optional tissue masking for reinhard (statistics computation within tissue regions).
• unittest updates.

V1.0.2 Release

Changelog:

• Optional stain matrices caching for both Normalizer and Augmentor.
• Add unit testing.
• Add documentation.

Installation from PyPI:

pip install torch-staintools

v0.0.4

What's new

• add stain augmentation (macenko/vahadane)
• add the support of stain matrix caching to avoid recomputation (currently in augmentors only). Cached can be dumped/loaded to/from local file system for future reuses.