ImageJ Active Segmentation platform: Parallel Engine for ASP/IJ

This project accelerates the convolution core of the Active Segmentation Plugin (ASP) in ImageJ by offloading its most compute-intensive pixel operations to parallel hardware using TornadoVM. The goal is to achieve significant speedups without altering existing functionality or behavior. The approach focuses on identifying CPU bottlenecks, converting them into accelerator-friendly kernels, and executing them through a reusable TornadoVM pipeline. A strict correctness-first methodology ensures all accelerated outputs match the CPU reference in terms of boundary handling and numerical accuracy before any performance gains are considered. By integrating a robust, reusable parallel execution layer with safe CPU fallback, this work transforms ASP’s performance while preserving its reliability—laying the foundation for scalable acceleration across multiple filters and larger image-processing workflows.