pymatting

Implements six established alpha matting methods like Closed Form and KNN, allowing developers to compare and choose the best approach for specific image types, as listed in the README.

Includes fast multithreaded KNN search and preconditioners such as incomplete Cholesky, which accelerate conjugate gradient descent for efficient processing, as highlighted in the key features.

Supports CuPy and PyOpenCL for foreground estimation tasks, leveraging GPU computing to speed up multi-level methods, though it's limited to foreground estimation only.

Features extensive documentation, benchmarks, and 89% test coverage, ensuring reliability and ease of use for research and production, as shown in the badges and testing section.