Fullseye: an explainable classical-vision operator library and workbench for Physical AI

Abstract

Fullseye is a numpy-native library of typed classical image-processing and geometric-vision operators (2-D and 3-D), with an evolutionary pipeline-design mode evaluated under held-out gates, a Physical-AI perception stack (stereo → depth → point cloud → 6-DoF pose), and an HDevelop-style IDE (Fullseye Studio). Every operator carries a machine-readable Markdown note (call contract, references, licence, version fingerprint) designed to double as a retrieval corpus for AI assistants and as a reproducible, citable methods reference. The repository also carries a series of worked proof-of-concept studies, each of which holds its ground truth in closed form or in synthesis and reports the null (do-nothing) result beside the method, so that a reader can see where a measurement breaks rather than only where it succeeds. Documentation: https://furuse.work/ · Distribution: https://pypi.org/project/fullseye/