DeepSDFStruct Documentation#

DeepSDFStruct is a differentiable framework for generating and deforming 3D microstructured materials using Signed Distance Functions (SDFs) and spline-based lattices. The library provides tools for creating, manipulating, and optimizing complex 3D geometries with applications in additive manufacturing, materials science, and computational design.

Overview#

DeepSDFStruct combines classical geometric representations (SDFs, splines) with deep learning approaches to enable:

  • Flexible Geometry Definition: Define complex 3D structures using primitive shapes, lattice patterns, or learned implicit representations

  • Differentiable Operations: All operations are PyTorch-compatible, enabling gradient-based optimization

  • Mesh Generation: Extract high-quality surface and volume meshes using FlexiCubes and FlexiSquares algorithms

  • Deep Learning Integration: Train neural networks to represent and generate novel geometric structures

  • Structural Optimization: Optimize designs for specific mechanical properties or manufacturing constraints

Key Features#

Signed Distance Functions (SDFs)#

The core abstraction is the SDFBase class, which represents geometries as implicit functions that return the signed distance to the nearest surface. The library provides:

  • Primitive Shapes: Spheres, cylinders, boxes, tori, and other basic geometric primitives

  • Lattice Structures: Periodic microstructures with customizable unit cells

  • Mesh-based SDFs: Convert existing triangle meshes to SDF representations

  • Deep Learning SDFs: Neural network-based implicit representations trained from data

  • Composition Operations: Combine multiple SDFs using boolean operations (union, intersection, difference)

Spline-based Deformations#

DeepSDFStruct uses B-splines for smooth, parametric deformations:

  • TorchSpline: PyTorch-compatible spline evaluation for differentiable operations

  • Parametrization: Define spatially-varying material properties or geometry parameters

Mesh Generation#

Generate high-quality meshes from SDF representations:

  • FlexiCubes: Advanced dual contouring for smooth, feature-preserving surface meshes

  • FlexiSquares: 2D mesh extraction for cross-sections and planar geometries

  • Volume Meshing: Generate tetrahedral meshes for finite element analysis

  • Mesh Processing: Clean, repair, and optimize generated meshes

Deep Learning Components#

Train neural networks to learn geometric representations:

  • DeepSDF Architecture: Multi-layer perceptron decoder for implicit geometry

  • Hierarchical Models: Multi-scale representations for complex structures

  • Training Pipeline: Complete workflow for dataset generation, training, and inference

  • Latent Space Optimization: Optimize learned representations for specific objectives

Optimization and Analysis#

Tools for structural design and optimization:

  • Gradient-based Optimization: Leverage automatic differentiation for shape optimization

  • MMA Optimizer: Method of Moving Asymptotes for constrained optimization

  • Finite Element Integration: Interface with TorchFEM for structural analysis

  • Design of Experiments: Generate parameter sweeps and design spaces

Core Modules#

  • DeepSDFStruct.SDF: Base classes for signed distance functions

  • DeepSDFStruct.mesh: Mesh generation and processing utilities

  • DeepSDFStruct.lattice_structure: Periodic lattice structure generation

  • DeepSDFStruct.sdf_primitives: Primitive geometric shapes

  • DeepSDFStruct.sampling: Strategies for sampling points from SDFs

  • DeepSDFStruct.optimization: Optimization algorithms and utilities

  • DeepSDFStruct.torch_spline: PyTorch-compatible B-spline operations

  • DeepSDFStruct.parametrization: Parametrization functions for spatially-varying properties

  • DeepSDFStruct.deep_sdf: Deep learning models and training pipelines

  • DeepSDFStruct.flexicubes: FlexiCubes mesh extraction algorithm

  • DeepSDFStruct.flexisquares: FlexiSquares 2D mesh extraction

  • DeepSDFStruct.splinepy_unitcells: Predefined unit cell geometries

Installation#

Install directly from GitHub using pip:

pip install git+https://github.com/mkofler96/DeepSDFStruct.git

Or add to your UV project:

uv add git+https://github.com/mkofler96/DeepSDFStruct.git

Quick Start#

Basic SDF operations:

import torch
from DeepSDFStruct.sdf_primitives import SphereSDF, CylinderSDF
from DeepSDFStruct.mesh import create_3D_mesh

# Create a sphere SDF
sphere = SphereSDF(center=[0, 0, 0], radius=0.5)

# Query SDF values
points = torch.tensor([[0.0, 0.0, 0.0], [1.0, 0.0, 0.0]])
distances = sphere(points)

# Generate mesh
mesh = create_3D_mesh(sphere, resolution=64)

For more examples, see the example notebook.

Contents:

Indices and tables#