S2MDF: A Plug-And-Play Layer for Intersection-Free Multi-Object Signed Distance Fields
Not on the map. Why?
Abstract (arXiv)
Compositional implicit surface representations model scenes as collections of objects, each encoded by a Signed Distance Field (SDF). A fundamental limitation of this approach is that multiple SDFs can produce geometries that interpenetrate, violating physical plausibility. Existing mitigation strategies rely on soft penalty terms that reduce but do not eliminate intersections, and require careful loss weighting. To truly prevent interpenetration, we propose a hard constraint on vector-valued SDFs and introduce S2MDF, a lightweight plug-and-play module that enforces the constraint on any object-compositional SDF representation without architectural modifications. It introduces negligible computational overhead and is compatible with linearly-interpolated standard meshing algorithms such as Marching Cubes. It can be applied during training or as a post-processing step. Experiments on multiple state-of-the-art compositional methods show that S2MDF reduces intersections to numerical precision while preserving reconstruction quality, outperforming existing mitigation strategies.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 55 of its 70 references that are in the map. First posted on arXiv 28 May 2026.
Builds on
55 of 70 references.
References · 55
- 2019DeepSDF: Learning Continuous Signed Distance Functions for Shape Representationcited by 2,689
- 2021Convex optimizationcited by 2,515
- 2018Occupancy Networks: Learning 3D Reconstruction in Function Spacecited by 2,041
- 2021NeuS: Learning Neural Implicit Surfaces by Volume Rendering for Multi-view Reconstructioncited by 1,712
- 2018Learning Implicit Fields for Generative Shape Modelingcited by 1,060
- 2016A Point Set Generation Network for 3D Object Reconstruction from a Single Imagecited by 979
- 2019The Replica Dataset: A Digital Replica of Indoor Spacescited by 904
- 2020GIRAFFE: Representing Scenes as Compositional Generative Neural Feature Fieldscited by 785
- 2018Pixel2Mesh: Generating 3D Mesh Models from Single RGB Imagescited by 651
- 2022TotalSegmentator: robust segmentation of 104 anatomical structures in CT imagescited by 613
- 2023Nerfstudio: A Modular Framework for Neural Radiance Field Developmentcited by 600
- 2017OptNet: Differentiable Optimization as a Layer in Neural Networkscited by 599
38 more
- 20163D-R2N2: A Unified Approach for Single and Multi-view 3D Object Reconstructioncited by 588
- 2016Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modelingcited by 558
- 2022LION: Latent Point Diffusion Models for 3D Shape Generationcited by 553
- 2018AtlasNet: A Papier-Mach'e Approach to Learning 3D Surface Generationcited by 539
- 2018PartNet: A Large-scale Benchmark for Fine-grained and Hierarchical Part-level 3D Object Understandingcited by 526
- 2022MonoSDF: Exploring Monocular Geometric Cues for Neural Implicit Surface Reconstructioncited by 470
- 2021Deep Marching Tetrahedra: a Hybrid Representation for High-Resolution 3D Shape Synthesiscited by 460
- 20233DShape2VecSet: A 3D Shape Representation for Neural Fields and Generative Diffusion Modelscited by 458
- 2024CLAY: A Controllable Large-scale Generative Model for Creating High-quality 3D Assetscited by 455
- 2021Neural Geometric Level of Detail: Real-time Rendering with Implicit 3D Shapescited by 426
- 2018Learning Category-Specific Mesh Reconstruction from Image Collectionscited by 317
- 2020Neural Unsigned Distance Fields for Implicit Function Learningcited by 265
- 2016Shape Completion using 3D-Encoder-Predictor CNNs and Shape Synthesiscited by 219
- 2022AutoSDF: Shape Priors for 3D Completion, Reconstruction and Generationcited by 208
- 2021Shape As Points: A Differentiable Poisson Solvercited by 187
- 2020Grasping Field: Learning Implicit Representations for Human Graspscited by 179
- 2019SDFDiff: Differentiable Rendering of Signed Distance Fields for 3D Shape Optimizationcited by 159
- 2025Hunyuan3D 2.5: Towards High-Fidelity 3D Assets Generation with Ultimate Detailscited by 125
- 2019PQ-NET: A Generative Part Seq2Seq Network for 3D Shapescited by 119
- 2020MeshSDF: Differentiable Iso-Surface Extractioncited by 112
- 2019Deep Mesh Reconstruction from Single RGB Images via Topology Modification Networkscited by 106
- 2024Dora: Sampling and Benchmarking for 3D Shape Variational Auto-Encoderscited by 98
- 20223DILG: Irregular Latent Grids for 3D Generative Modelingcited by 93
- 2020DualSDF: Semantic Shape Manipulation using a Two-Level Representationcited by 82
- 2022Object-Compositional Neural Implicit Surfacescited by 74
- 2022DM-NeRF: 3D Scene Geometry Decomposition and Manipulation from 2D Imagescited by 69
- 2023Ref-NeuS: Ambiguity-Reduced Neural Implicit Surface Learning for Multi-View Reconstruction with Reflectioncited by 69
- 2023vMAP: Vectorised Object Mapping for Neural Field SLAMcited by 67
- 2022NeuralUDF: Learning Unsigned Distance Fields for Multi-view Reconstruction of Surfaces with Arbitrary Topologiescited by 63
- 2021FiG-NeRF: Figure-Ground Neural Radiance Fields for 3D Object Category Modellingcited by 47
- 2023NeUDF: Leaning Neural Unsigned Distance Fields with Volume Renderingcited by 41
- 2023ObjectSDF++: Improved Object-Compositional Neural Implicit Surfacescited by 39
- 2023PartNeRF: Generating Part-Aware Editable 3D Shapes without 3D Supervisioncited by 36
- 2023Surface Extraction from Neural Unsigned Distance Fieldscited by 17
- 2022ANISE: Assembly-based Neural Implicit Surface rEconstructioncited by 16
- 2022An Implicit Parametric Morphable Dental Modelcited by 12
- 2025PartSDF: Part-Based Implicit Neural Representation for Composite 3D Shape Parametrization and Optimizationcited by 9
- 2024MedTet: An Online Motion Model for 4D Heart Reconstructioncited by 5
50 most cited of 55 All 55 (JSON)
Built on by
Citing papers per year: 2026: 1.
1 paper.