Additive Growth Generation: A Constructive Framework for 3D Object Synthesis Through Adaptive Incremental Unit Deposition
Abstract
Most three-dimensional generative models represent object synthesis as the sampling or optimization of a global shape representation, such as a point cloud, voxel grid, polygonal mesh, neural implicit field, or collection of three-dimensional primitives. This paper proposes Additive Growth Generation (AGG), a constructive framework in which a three-dimensional object emerges through a sequence of explicit addition operations. Rather than requiring the complete geometry to be generated simultaneously, AGG maintains an evolving partial object and repeatedly selects an addition action, including the type, position, orientation, scale, attributes, and attachment configuration of a candidate unit. A learned growth controller conditions these decisions on the current partial geometry, the target specification, local structural context, and global completion requirements. The framework supports variable-cardinality representations, multiple unit granularities, and alternative geometric decoders. Its central formulation separates the decision to add a unit from the representation used to encode the resulting object. This separation enables the same constructive process to operate over particles, surface elements, volumetric cells, geometric primitives, or larger structural components. We define a probabilistic transition model, a multi-objective training formulation, structural validity constraints, and an evaluation protocol for geometric fidelity, growth efficiency, structural consistency, and controllability. Potential applications include generative design, crystal-like structures, additive manufacturing, and procedurally constructed objects. This manuscript presents a proposed framework rather than a validated empirical result. Its novelty and effectiveness must be established through implementation, comparative experiments, and further investigation of related constructive generation methods.
Related
- Granular Deposition Image Generation: A Framework for Target-Guided Visual Synthesis Through Particle Transport, Selective Deposition, and Dynamic Reorganization
- River-Erosion-Based Image Generation: A Feedback-Driven Framework for Iterative Visual Synthesis Through Erosion, Transport, and Deposition
- Adaptive Layered Deposition for Image Generation: A State-Dependent Framework for Progressive Construction, Selective Modification, and Variable-Depth Synthesis
- Layerwise Deposition-Field Generative Modeling: A Process-Conditioned Framework for Three-Dimensional Generation Through Thin-Film Deposition and Surface Growth
- Deposition-Based 3D Generative Modeling: A Framework for Three-Dimensional Object Generation Through Gas-Phase Particle Deposition, Local Adsorption, and Branching Growth
BibTeX
@misc{zhang2026additive,
title = {{Additive Growth Generation: A Constructive Framework for 3D Object Synthesis Through Adaptive Incremental Unit Deposition}},
author = {Jincheng Zhang},
howpublished = {Zenodo},
year = {2026},
month = oct,
doi = {10.5281/zenodo.23636364},
url = {https://zenodo.org/records/23636364},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Additive Growth Generation: A Constructive Framework for 3D Object Synthesis Through Adaptive Incremental Unit Deposition
AU - Jincheng Zhang
PY - 2026
DA - 2026/10/25
PB - Zenodo
DO - 10.5281/zenodo.23636364
UR - https://zenodo.org/records/23636364
AB - Most three-dimensional generative models represent object synthesis as the sampling or optimization of a global shape representation, such as a point cloud, voxel grid, polygonal mesh, neural implicit field, or collection of three-dimensional primitives. This paper proposes Additive Growth Generation (AGG), a constructive framework in which a three-dimensional object emerges through a sequence of explicit addition operations. Rather than requiring the complete geometry to be generated simultaneously, AGG maintains an evolving partial object and repeatedly selects an addition action, including the type, position, orientation, scale, attributes, and attachment configuration of a candidate unit. A learned growth controller conditions these decisions on the current partial geometry, the target specification, local structural context, and global completion requirements. The framework supports variable-cardinality representations, multiple unit granularities, and alternative geometric decoders. Its central formulation separates the decision to add a unit from the representation used to encode the resulting object. This separation enables the same constructive process to operate over particles, surface elements, volumetric cells, geometric primitives, or larger structural components. We define a probabilistic transition model, a multi-objective training formulation, structural validity constraints, and an evaluation protocol for geometric fidelity, growth efficiency, structural consistency, and controllability. Potential applications include generative design, crystal-like structures, additive manufacturing, and procedurally constructed objects. This manuscript presents a proposed framework rather than a validated empirical result. Its novelty and effectiveness must be established through implementation, comparative experiments, and further investigation of related constructive generation methods.
ER -