tgLang: A Domain-Specific Language for Geometry Processing and Computational Imaging Workflows

Abstract

Geometry-processing and computational-imaging workflows combine heterogeneous data structures, topology-changing edits, dense numerical fields, visualization, and repeated experimental variation. These workflows are often clear as algorithms but obscured in software by traversal boilerplate, representation conversions, build-system boundaries, and ad hoc scripting conventions. This paper presents tgLang, a domain-specific language with explicit, runtime-enforced representation types that makes meshes, point clouds, curve networks, grids, two-dimensional images, and image stacks first-class executable values. The language combines manifest types, typed arrays, modules, deterministic parallel constructs, flow-oriented queries, and runtime-provided domain operations. Its current implementation uses a stack-based bytecode virtual machine for reference semantics and dispatches representation-heavy operations to optimized C++ kernels. The evaluation is organized around complete workflows: topological hole detection, distance-field-based mean camber line extraction, voxel downsampling of point clouds, curve-network generation, surface-mesh smoothing and remeshing, image-stack edge detection, morphological image processing, and image-stack surface extraction. These examples show that a domain-aware source language can express multi-representation geometry and imaging algorithms as compact, reproducible programs while preserving explicit representation choices and a path toward deployable implementations.