Deletion-Driven 3D Generation: A Progressive Removal Framework for Shape, Topology, and Material-Aware Three-Dimensional Synthesis
Abstract
Most three-dimensional (3D) generative methods represent shape synthesis through the prediction, transformation, or refinement of geometric representations. This paper proposes Deletion-Driven 3D Generation (DD3G), a conceptual framework that formulates 3D shape synthesis as a progressive removal process. Instead of constructing a target object primarily by adding geometric primitives or predicting its final representation directly, DD3G begins with an initial volumetric structure and iteratively removes material, particles, or structural elements that are inconsistent with a target specification. The framework integrates target-conditioned removal decisions, geometry-aware candidate selection, topology monitoring, material-retention constraints, and removal-cost optimization. A unified mathematical formulation is developed for voxel, point-based, mesh-based, and implicit representations. The proposed objective balances target-shape fidelity, structural and functional preservation, removal cost, and manufacturing feasibility. The framework further distinguishes discrete deletion from continuous erosion and selective dissolution, allowing different removal mechanisms to be modeled within a common decision process. A proposed learning strategy combines supervised removal prediction with sequential decision optimization and geometric feedback. Potential applications include virtual sculpting, subtractive manufacturing, erosion-inspired modeling, selective dissolution simulation, and the generation of lightweight 3D structures. This paper establishes the conceptual and mathematical foundations of DD3G and presents a research protocol for evaluating it against existing generative and optimization methods. No experimental results are claimed. The principal research hypothesis is that explicitly modeling removal as a generative action may improve controllability and material-efficiency awareness in tasks where the desired object is naturally described as a retained subset of an initial structure.
Related
- Adaptive Layered Deposition for Image Generation: A State-Dependent Framework for Progressive Construction, Selective Modification, and Variable-Depth Synthesis
- Wear-Driven Generative Modeling: A Conditional Framework for Three-Dimensional Shape Synthesis Through Progressive Material Removal
- Splitting-and-Branching 3D Generation: A Recursive Structural Growth Framework for Three-Dimensional Shape Synthesis
- Phase Redistribution and Separation Generation: A Component-Transport Framework for Controllable Three-Dimensional Microstructure Synthesis
- Gas-Expansion and Lock-In Synthesis: A Process-Guided Framework for Three-Dimensional Generation of Cellular and Porous Structures
BibTeX
@misc{zhang2026deletiondriven,
title = {{Deletion-Driven 3D Generation: A Progressive Removal Framework for Shape, Topology, and Material-Aware Three-Dimensional Synthesis}},
author = {Jincheng Zhang},
howpublished = {Zenodo},
year = {2026},
month = oct,
doi = {10.5281/zenodo.23636388},
url = {https://zenodo.org/records/23636388},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Deletion-Driven 3D Generation: A Progressive Removal Framework for Shape, Topology, and Material-Aware Three-Dimensional Synthesis
AU - Jincheng Zhang
PY - 2026
DA - 2026/10/25
PB - Zenodo
DO - 10.5281/zenodo.23636388
UR - https://zenodo.org/records/23636388
AB - Most three-dimensional (3D) generative methods represent shape synthesis through the prediction, transformation, or refinement of geometric representations. This paper proposes Deletion-Driven 3D Generation (DD3G), a conceptual framework that formulates 3D shape synthesis as a progressive removal process. Instead of constructing a target object primarily by adding geometric primitives or predicting its final representation directly, DD3G begins with an initial volumetric structure and iteratively removes material, particles, or structural elements that are inconsistent with a target specification. The framework integrates target-conditioned removal decisions, geometry-aware candidate selection, topology monitoring, material-retention constraints, and removal-cost optimization. A unified mathematical formulation is developed for voxel, point-based, mesh-based, and implicit representations. The proposed objective balances target-shape fidelity, structural and functional preservation, removal cost, and manufacturing feasibility. The framework further distinguishes discrete deletion from continuous erosion and selective dissolution, allowing different removal mechanisms to be modeled within a common decision process. A proposed learning strategy combines supervised removal prediction with sequential decision optimization and geometric feedback. Potential applications include virtual sculpting, subtractive manufacturing, erosion-inspired modeling, selective dissolution simulation, and the generation of lightweight 3D structures. This paper establishes the conceptual and mathematical foundations of DD3G and presents a research protocol for evaluating it against existing generative and optimization methods. No experimental results are claimed. The principal research hypothesis is that explicitly modeling removal as a generative action may improve controllability and material-efficiency awareness in tasks where the desired object is naturally described as a retained subset of an initial structure.
ER -