Computational Frameworks for Geometric Hollowing in Additive Manufacturing CAD Software

The evolution of additive manufacturing (AM) from a niche prototyping tool to a primary industrial fabrication method has necessitated a paradigm shift in geometric processing. At the core of this transformation is the requirement for internal hollowing and shelling—operations that serve the dual purpose of material reduction and functional enhancement. As geometries become increasingly complex, moving from simple mechanical parts to organic, high-genus structures, the limitations of traditional Boundary Representation (B-rep) kernels have become evident. Modern hollowing algorithms must now operate with a high degree of topological awareness, ensuring that the resulting shells are not only visually accurate but also structurally sound, printable, and mathematically manifold. This report provides an exhaustive investigation into state-of-the-art algorithms, from implicit volumetric representations and homogenization-based lattice design to the fluid dynamic considerations of VAT polymerization and the skeletal logic of the Medial Axis Transform.

Theoretical Foundations of Geometric Shelling and Volumetric Representations

The fundamental challenge in hollowing for additive manufacturing is the generation of an internal offset surface that remains equidistant from the exterior boundary while avoiding topological failures such as self-intersection and non-manifoldness. Historically, this was approached through direct manipulation of mesh vertices along their normal vectors. However, for complex geometries—those with sharp features, high curvature, or narrow passages—this explicit approach often leads to “explosions” in geometry where facets overlap or intersect. The contemporary shift toward volumetric and implicit methods reflects a need for robustness in the face of these geometric singularities.

Voxel-Based Hollowing and Signed Distance Fields (SDF)

Voxel-based hollowing represents the transition from a surface-centric view to a volumetric one. By embedding a 3D model into a discrete grid of voxels, geometric operations can be performed as field calculations rather than combinatorial mesh edits. The most robust implementation of this is the Signed Distance Field (SDF). An SDF is a scalar field where each point in the design domain is assigned a value representing the Euclidean distance to the nearest point on the boundary . The sign of the value indicates whether the point is inside (negative) or outside (positive) the solid.

For a hollow part with a desired wall thickness , the internal surface is defined as the zero-level set of the modified distance function:

This mathematical formulation naturally handles complex topological changes. Because the hollowing is performed in a distance field, self-intersections are inherently impossible; the field simply converges at the medial axis, creating a clean internal boundary regardless of external complexity.1 To achieve efficiency in high-resolution grids, narrow-band SDF sampling is often employed, where the distance field is only calculated within a specific range of the surface, significantly reducing the computational overhead.3

Explicit Mesh Offsetting and Minkowski Sums

In contrast to implicit fields, explicit hollowing attempts to construct the offset boundary directly. The Minkowski sum provides the theoretical framework for this operation. For two sets of points and in , their Minkowski sum is defined as . If is the input geometry and is a sphere of radius , the boundary of their Minkowski sum represents the offset surface. While mathematically precise, the exact computation for arbitrary polyhedra is computationally intensive and plagued by numerical instability.3

The primary issue with explicit Minkowski-based hollowing is the “self-intersection” problem. When an offset distance is greater than the local radius of curvature on a concave feature, the resulting triangles will cross each other. Resolving these requires complex boolean operations and re-meshing, which are often the bottleneck of a geometric kernel.4

 

Feature	Voxel-Based SDF Hollowing	Explicit Minkowski Offsetting
Representational Method	Implicit (Volumetric Field)	Explicit (Boundary Mesh)
Topological Robustness	Inherently handles self-intersections 1	Prone to non-manifold overlaps 4
Computational Efficiency	with parallel Fast Sweeping 1	or higher for complex cases 3
Feature Preservation	Limited by grid resolution 2	High (preserves original vertices) 6
Manifoldness Guarantee	Guaranteed with Dual Contouring 2	Requires extensive post-repair 5
Memory Consumption	High (Octrees required to mitigate) 2	Lower (Scales with surface area) 6

Despite the advantages of SDFs, they introduce a discretization error. Recent research like “OffsetCrust” attempts to combine the two paradigms by using power diagrams to compute variable-radius offsets that remain manifold while preserving the fidelity of the explicit representation.6

Self-Intersection Prevention in Complex Geometries

The prevention of self-intersections is the hallmark of a “production-ready” hollowing algorithm. In explicit systems, this is typically handled through a combination of local vertex pruning and global re-meshing. However, the most robust prevention mechanism is the use of a volumetric octree data structure to compute a distance field.2

Implicit Surface Level Sets

By representing the shell as a level set of an implicit function, self-intersections are resolved because the function is single-valued at every point in space. If two parts of the shell were to “intersect,” the distance field simply identifies the shortest path to either surface, naturally forming a medial ridge (the Voronoi bisector) between them.4 This property is particularly valuable for high-genus models from datasets like Thingi10K, where traditional B-rep kernels frequently fail.6

Dual Contouring and Feature Preservation

Once the internal volume is defined via SDF, the mesh must be extracted. Traditional Marching Cubes often fail to capture sharp edges and can produce non-manifold geometry in thin regions. Dual Contouring (DC) addresses this by placing vertices based on a Quadratic Error Function (QEF):

where and are the intersection points and normals on the grid edges. By minimizing this error, the algorithm positions the vertex at the intersection of the tangent planes of the original surface, effectively preserving sharp corners and preventing the “aliasing” artifacts common in voxelization.4 To ensure manifold output, modern DC implementations incorporate a vertex clustering scheme and topological checks based on Topologically-Correct Marching Cubes (TMC) rules.2

Mathematical Methods for Internal Lattice Structures

Hollowing reduces the mass of a part, but often at the cost of its structural integrity. To counteract this, hollowing pipelines often integrate internal lattice structures. These are typically generated through mathematical functions that define periodic architectures.

Triply Periodic Minimal Surfaces (TPMS)

TPMS lattices are favored in modern AM because they provide a continuous, smooth material distribution that minimizes stress concentrations at nodal joints. They are defined by implicit functions of the form .

 

Lattice Type	Governing Function Φ(x,y,z)=C	Symmetry and Application
Schwarz P		Cubic; High surface area-to-volume ratio 8
Gyroid		Isometric; Excellent energy absorption 8
Diamond		Highly anisotropic at low densities 8
I-WP		High connectivity; Complex labyrinth 8

The Gyroid, in particular, is noted for its high degree of isotropy even at low relative densities (), making it a candidate for parts subjected to multi-axial loading.9 The wall thickness of these lattices is controlled by the offset , which can be spatially varied to create functionally graded materials.10

Truss-Based Lattices

Truss-based lattices are composed of repeating unit cells such as tetrahedrons, octahedrons, and dodecahedrons.12 Unlike TPMS, these are typically designed through 3D texture mapping, where a wireframe unit cell is mapped into a design space and combined with the shell via Boolean operations. While computationally simpler, they are prone to stress concentrations at the junctions where struts meet.12

Homogenization Theory for Lattice Optimization

The computational expense of simulating a full-scale part filled with millions of lattice unit cells is immense. Homogenization theory provides a multiscale approach to this problem by calculating the “effective properties” of a lattice unit cell and treating the entire lattice as a continuous, equivalent solid.

Representative Volume Element (RVE) Analysis

Homogenization begins with the definition of an RVE or Representative Unit Cell (RUC). By applying periodic boundary conditions and solving a series of local elastic problems, the effective stiffness matrix of the lattice is derived.13 This matrix represents how the lattice behaves at a macroscale. For low-density truss lattices (), Euler-Bernoulli beam theory is often sufficient for these calculations.13

The HOC Framework: Homogenization, Optimization, and Construction

The most advanced lattice generation pipelines use the HOC framework 12:

1. Homogenization: Determine effective properties (Young’s modulus, Poisson’s ratio) for various lattice topologies and densities.
2. Optimization: Use continuous topology optimization (e.g., SIMP) to solve the macroscale problem, determining where the lattice should be dense or sparse.16
3. Construction: Map the optimized density field back to the lattice parameters to generate a non-uniform, stress-optimized internal structure.

Recent studies comparing Finite Element (FE)-based homogenization with the Mechanics of Structure Genome (MSG) have shown that MSG can achieve the same accuracy (less than 1% difference) while being significantly more efficient, particularly for anisotropic TPMS structures.9

Fluid Dynamics and Pressure Differentials in VAT Polymerization

While geometric hollowing is mathematically elegant, its physical implementation in VAT polymerization (SLA, DLP, LCD) introduces unique challenges related to fluid dynamics. The two primary failure modes are “suction cupping” and “resin entrapment.”

The Suction Cupping Phenomenon (Blowout)

In bottom-up SLA, each layer is cured against a flexible film (FEP) at the bottom of a resin tank and then “peeled” away. If the part has an enclosed cavity or a cup-like geometry facing the build platform, it traps air or liquid resin.18 As the platform lifts, the trapped volume expands, creating a negative pressure differential. This pressure differential exerts an inward force on the part’s walls. If the wall thickness is insufficient to withstand this stress, the part may rupture—an event known as a “blowout”.19

The magnitude of this force is governed by the resin’s viscosity and the lifting speed. Experimental data shows that as resin viscosity increases from 0.05 to 1 Pa.s, the stabilization time for the fluid interface increases nearly fourfold, intensifying the suction effect.22

Stefan Formula and Peel Forces

The peel force generated during layer separation can be approximated using the Stefan formula for viscous fluids between two parallel plates:

where is viscosity, is lift speed, is the radius of the layer, and is the distance to the film.24 This equation demonstrates a critical insight: the force scales with . Thus, even a small increase in the hollow’s cross-sectional area leads to an exponential increase in print stress. Managing these forces requires slowing the lift speed or, more effectively, adding drainage holes.24

 

Parameter	Recommended Specification	Impact on Print Success
Min. Drainage Hole Diameter	3.5 mm 26	Prevents air-lock and ensures resin flow.
Min. Vent Hole Diameter	1.3 – 2.5 mm 28	Relieves vacuum pressure during peel.
Hollow Wall Thickness	mm 26	Provides structural rigidity against suction.
Orientation Angle	26	Minimizes surface area per layer and suction.
Resin Viscosity	140 – 550 cP 23	Influences peel forces and stabilization time.

Algorithms for Optimal Drainage Hole Placement

Optimal drainage hole placement is a problem of both geometry and fluid mechanics. Standard slicers often rely on manual placement or simple heuristic-based automated tools, but state-of-the-art algorithms are now employing path-based search and pressure field simulation.

Local Minima Search and Pathfinding

For resin drainage, holes must be placed at the local minima (the lowest points relative to the ground) of every hollow section. This ensures that gravity can pull the uncured resin out of the finished part.20 Conversely, for suction relief, a hole should be placed as close as possible to the build platform (the first layer of the cavity) to break the vacuum immediately.20 A robust placement algorithm performs a segmentation of the hollowed volume and uses a connectivity-based search to ensure that every “cup” has at least one path to the external environment.26

Multiple Hole Strategy and Air-Lock Prevention

Advanced research emphasizes the “two-hole rule”: one hole for drainage and a second hole (vent) to allow air to enter the cavity as the resin exits.29 Without a vent, the surface tension and negative pressure within the hollow can cause an air-lock, trapping resin even if a drainage hole is present. The optimal algorithm identifies the gravity vector and places a drain at the global minimum and a vent at the global maximum of the internal void.20

Medial Axis Transform (MAT) for Variable-Thickness Hollowing

While constant-thickness shelling is standard, it is rarely optimal from a structural perspective. The Medial Axis Transform (MAT) offers a skeletal representation that allows for adaptive, stress-informed wall thickness.

Geometric and Topological Properties of MAT

The medial axis of a shape is defined as the set of centers of maximal inscribed spheres that are tangent to the boundary at two or more points.34 The MAT consists of the medial axis and the associated radius function for each point . Because the MAT captures the “local thickness” of the shape, it serves as a natural backbone for hollowing.36

MAT-based hollowing operates by defining the internal boundary as a function of the radius . For a hollow part, the internal surface is reconstructed from a “shrunk” medial axis where the radius is reduced by the desired wall thickness .

Variable Thickness and Structural Integrity

In high-stress regions, the wall thickness can be increased by reducing the offset distance from the medial axis. Conversely, in low-stress regions, the part can be hollowed more aggressively. This mapping is often informed by Finite Element Analysis (FEA) performed on a simplified skeletal model.38

The mathematical logic for MAT-driven hollowing involves deforming the medial axis in an “as-rigid-as-possible” (ARAP) manner, ensuring that the local transforms of the medial primitives (spheres and cones) remain consistent with the original geometry.39 To maintain structural integrity, hollowing, thickening, and strut insertion are used as corrective measures in regions where the structural solver detects potential failure.38

 

Metric	MAT-Based Hollowing	Constant-Thickness Hollowing
Material Efficiency	Optimized for stress distribution 16	Uniform; often over-engineered 40
Symmetry Preservation	High; follows the geometric backbone 41	Moderate; independent of local skeleton
Computational Cost	High (MAT extraction is expensive) 42	Low to Moderate (Simple offset)
Robustness	Sensitive to boundary noise 39	Highly robust in implicit domains 2
Feature Sensitivity	Captures protrusions and limbs naturally 35	May “pinch” in narrow regions

Recent advancements in MAT computation, such as the use of Restricted Power Diagrams (RPD), have significantly improved the stability and mesh quality of the medial structure, making it more viable for CAD-to-AM pipelines.34

A Robust Hollowing Pipeline: Mathematical Logic and Pseudo-code

To ensure a manifold, print-ready output, a hollowing pipeline must synthesize geometric, skeletal, and fluid-dynamic requirements into a single cohesive process. The following mathematical logic defines a robust pipeline designed to guarantee a 2-manifold output.

1. Mesh Validation and Manifold Repair

The input mesh is analyzed for topological consistency. A mesh is manifold if every edge is shared by exactly two faces and the neighborhood of every vertex is a single disk.44

• Logic: If non-manifold features are found, apply a vertex-clustering or bridge-edge removal algorithm to ensure represents a closed solid.44

2. SDF Generation via Adaptive Octree

Convert into a Signed Distance Field using an adaptive octree to minimize memory usage in large void regions.2

• Computation: Use the parallel Fast Sweeping Method to solve the Eikonal equation with boundary conditions on .1

3. Variable Thickness Mapping (MAT Integration)

If variable thickness is required, compute the Medial Axis and its radius function . Define the thickness field based on stress data .38

• Logic: . This generates the internal offset field.

4. Lattice Infill Synthesis

Evaluate the TPMS function (e.g., Gyroid) for the domain where .

• Logic: . The “max” operation in SDF space performs a robust Boolean union between the shell and the lattice.47

5. Automated Drainage and Suction Relief

Identify internal cavities by segmenting the regions where .

• Algorithm: For each cavity :

1. Find where gravity vector is minimized (Drain).
2. Find where is maximized (Vent).
3. Create cylinder SDFs and at these locations.
4. Update field: .20

6. Manifold Isosurface Extraction

Extract the final boundary mesh from using Manifold Dual Contouring.

• Constraint: For each octree cell intersecting the isosurface, compute the vertex by minimizing the QEF. Resolve “tunnel” ambiguities using TMC lookup tables to ensure a 2-manifold topology.2

7. Global Verification

Perform a final manifold check on and ensure minimum wall thickness is maintained for printability.26

Conclusions and Future Trajectories

The design of hollowing and shelling algorithms for additive manufacturing has matured from basic geometric offsetting to a multi-physics optimization problem. The integration of Signed Distance Fields has provided a robust solution to the self-intersection problem, while homogenization theory has unlocked the ability to use complex lattices as functional infills. However, as the industry moves toward “true” manufacturing, the physical constraints of the print process—particularly the fluid dynamics of resin systems—must be integrated directly into the geometric kernel.

The Medial Axis Transform remains the most promising skeletal representation for achieving biomimetic, variable-thickness shells, though its computational sensitivity necessitates further research into simplified, structure-aware medial meshes. The future of the field lies in “Self-Supervised Dual Contouring” and neural implicit representations, where machine learning models learn to extract high-quality, feature-preserving meshes that are inherently manifold and optimized for the specific constraints of the printing technology.49 By unifying these disparate disciplines—topology optimization, computational geometry, and fluid mechanics—the next generation of hollowing pipelines will enable the fabrication of parts with unprecedented efficiency and performance.

Works cited

1. Medial axis and local thickness computation using the Fast Sweeping Method – OSTI, accessed April 2, 2026, https://www.osti.gov/biblio/2406675
2. Evaluation of Manifold Dual Contouring Algorithms Based on k-d tree and Octree Data Structures 1 Introduction – Informatica, accessed April 2, 2026, https://www.informatica.si/index.php/informatica/article/download/5420/3244/0
3. Covering Minkowski sum boundary using points with applications – ResearchGate, accessed April 2, 2026, https://www.researchgate.net/publication/222415274_Covering_Minkowski_sum_boundary_using_points_with_applications
4. Non-uniform offsetting of surfaces, accessed April 2, 2026, https://www.cg.tuwien.ac.at/research/publications/2025/weydemann-2025-noo/weydemann-2025-noo-thesis.pdf
5. Mesh Offsetting · Issue #192 · elalish/manifold – GitHub, accessed April 2, 2026, https://github.com/elalish/manifold/issues/192
6. Thingi10K: A Dataset of 10,000 3D-Printing Models – ResearchGate, accessed April 2, 2026, https://www.researchgate.net/publication/303269016_Thingi10K_A_Dataset_of_10000_3D-Printing_Models
7. Enhanced Dual Contouring · Issue #9047 · CGAL/cgal – GitHub, accessed April 2, 2026, https://github.com/CGAL/cgal/issues/9047
8. Topological and Mechanical Properties of Different Lattice Structures Based on Additive Manufacturing – PMC, accessed April 2, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC9324303/
9. Mechanical Homogenisation of TPMS Architectures: A Comparison …, accessed April 2, 2026, https://www.mdpi.com/1996-1944/18/23/5357
10. Homogenization-based optimization of wall thickness distribution for TPMS two-fluid heat exchangers – arXiv, accessed April 2, 2026, https://arxiv.org/html/2510.10622v1
11. Dual Graded Lattice Structures: Generation Framework and Mechanical Properties Characterization – MDPI, accessed April 2, 2026, https://www.mdpi.com/2073-4360/13/9/1528
12. Design and Optimization of Lattice Structures: A Review – MDPI, accessed April 2, 2026, https://www.mdpi.com/2076-3417/10/18/6374
13. Homogenization Methods of Lattice Materials – MDPI, accessed April 2, 2026, https://www.mdpi.com/2673-8392/2/2/72
14. Consistent Asymptotic Homogenization Method for Lattice Structures Based on the Virtual Power Principle – Clemson OPEN, accessed April 2, 2026, https://open.clemson.edu/cgi/viewcontent.cgi?article=3552&context=all_theses
15. Status and Challenges in Homogenization Methods for Lattice Materials – PMC, accessed April 2, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC8778170/
16. Stress-Responsive Spatial Voronoi Optimization for Lightweight Architectural Shell Structures – MDPI, accessed April 2, 2026, https://www.mdpi.com/2075-5309/15/9/1547
17. Stress‐Constrained Thickness Optimization for Shell Object Fabrication | Request PDF – ResearchGate, accessed April 2, 2026, https://www.researchgate.net/publication/308608646_Stress-Constrained_Thickness_Optimization_for_Shell_Object_Fabrication
18. Preventing suction cups in PreForm – Support | Formlabs, accessed April 2, 2026, https://support.formlabs.com/s/article/Preventing-suction-cups-in-PreForm
19. Cupping blowout – Support | Formlabs, accessed April 2, 2026, https://support.formlabs.com/s/article/Cupping-Blowout
20. Moving from FDM to SLA: Part 1 – Formlabs, accessed April 2, 2026, https://formlabs.com/blog/how-to-orient-sla-parts/
21. SLA 3D Printing Design Guidelines: 8 Essential Tips | Xometry Pro, accessed April 2, 2026, https://xometry.pro/en-eu/articles/3d-printing-sla-design-tips/
22. Computational Fluid Dynamics Modeling of Top-Down Digital Light Processing Additive Manufacturing – PMC, accessed April 2, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC10255172/
23. Analysis and Case Study Vat Photopolymerization Additive Manufacturing Resins – SciELO, accessed April 2, 2026, https://www.scielo.br/j/mr/a/vZLwGmdnzSMhkT96J4PVNQx/?lang=en
24. How to Eliminate Z-axis Lines Caused by Peel/Separation Force in …, accessed April 2, 2026, https://www.instructables.com/How-to-Eliminate-Z-axis-Lines-Caused-by-Peel-Force/
25. How to Control Peel Forces & Prevent the Vacuum Effect in Resin 3D Printing, accessed April 2, 2026, https://protoresins.com/blog/control-peel-forces-prevent-the-vacuum-effect-in-resin-3d-printing
26. SLA 3D Printing Design Guidelines: 8 Essential Tips | Xometry Pro, accessed April 2, 2026, https://xometry.pro/en/articles/3d-printing-sla-design-tips/
27. Design Guide: SLA 3D Printing | Stereolithography – HLH Rapid, accessed April 2, 2026, https://hlhrapid.com/knowledge/design-guide-sla-3d-printing/
28. SLA Best Practices Quickcast Post Processing Guide Draining – 3D Systems Support Home, accessed April 2, 2026, https://support.3dsystems.com/s/article/sla-best-practices-draining-qcast
29. An operational guide to resin 3D printing of geological macromodels – PMC – NIH, accessed April 2, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC9513601/
30. Vat Photopolymerization Additive Manufacturing Resins: Analysis and Case Study – SciSpace, accessed April 2, 2026, https://scispace.com/pdf/vat-photopolymerization-additive-manufacturing-resins-4shhz7z46t.pdf
31. Resin Printing Tips – Drainage Holes Explained – YouTube, accessed April 2, 2026, https://www.youtube.com/watch?v=Me8I_wiKMyI
32. Model orientation best practices for SLA printing – Support | Formlabs, accessed April 2, 2026, https://support.formlabs.com/s/article/Model-Orientation?language=en_US
33. Trapped volume – onsite, accessed April 2, 2026, https://onsite.helpjuice.com/design-printing/trapped-volume
34. High-Quality Medial Mesh Computation via Structure-aware Variational Optimization – arXiv, accessed April 2, 2026, https://arxiv.org/html/2510.10751v1
35. Computing Medial Axis Transform with Feature Preservation via Restricted Power Diagram – The University of Texas at Dallas, accessed April 2, 2026, https://www.utdallas.edu/~xguo/SA2022.pdf
36. (PDF) Medial axis tree – An internal supporting structure for 3D printing – ResearchGate, accessed April 2, 2026, https://www.researchgate.net/publication/274096061_Medial_axis_tree_-_An_internal_supporting_structure_for_3D_printing
37. Stability and Computation of Medial Axes — a State-of-the-Art Report, accessed April 2, 2026, https://pub.ista.ac.at/~edels/Papers/2009-01-MedialAxis.pdf
38. Stress Relief: Improving Structural Strength of 3D Printable Objects – CS@Purdue, accessed April 2, 2026, https://www.cs.purdue.edu/cgvlab/www/resources/papers/Stava-ACM_Trans_Graph-2012-Stress_Relief_Improving_Structural_Strength_of_3D_Printable_Obje.pdf
39. Medial-Axis-Driven Shape Deformation with Volume Preservation – The University of Texas at Dallas, accessed April 2, 2026, https://www.utdallas.edu/~xxg061000/CGI17.pdf
40. (PDF) Topology Optimization for Variable Thickness Shell-Infill Composites Based on Stress Analysis Preprocessing – ResearchGate, accessed April 2, 2026, https://www.researchgate.net/publication/394369744_Topology_Optimization_for_Variable_Thickness_Shell-Infill_Composites_Based_on_Stress_Analysis_Preprocessing
41. hex-dominant mesh generation: Topics by Science.gov, accessed April 2, 2026, https://www.science.gov/topicpages/h/hex-dominant+mesh+generation
42. Medial Meshes – A Compact and Accurate Representation of Medial Axis Transform – HKU, accessed April 2, 2026, https://i.cs.hku.hk/~yzyu/publication/medialmesh-tvcg2016.pdf
43. Medial Skeletal Diagram: A Generalized Medial Axis Approach for 3D Shape Representation – arXiv, accessed April 2, 2026, https://arxiv.org/html/2310.09395v3
44. elalish/manifold: Geometry library for topological robustness – GitHub, accessed April 2, 2026, https://github.com/elalish/manifold
45. Manifold Library · elalish/manifold Wiki – GitHub, accessed April 2, 2026, https://github.com/elalish/manifold/wiki/Manifold-Library/ee3eee76533de1223af38a45e1bc430223c71af1
46. Manifold Library · elalish/manifold Wiki – GitHub, accessed April 2, 2026, https://github.com/elalish/manifold/wiki/Manifold-Library
47. Redundant SDFs – SGI 2024, accessed April 2, 2026, https://summergeometry.org/sgi2024/redundant-sdfs/
48. SDF optimizations · elalish manifold · Discussion #561 – GitHub, accessed April 2, 2026, https://github.com/elalish/manifold/discussions/561
49. Self-Supervised Dual Contouring – CVF Open Access, accessed April 2, 2026, https://openaccess.thecvf.com/content/CVPR2024/papers/Sundararaman_Self-Supervised_Dual_Contouring_CVPR_2024_paper.pdf
50. Self-Supervised Dual Contouring – arXiv, accessed April 2, 2026, https://arxiv.org/html/2405.18131v1