Open research questions in 3D Shape Modeling and Analysis

What the literature leaves open

• To address current limitations, future work will focus on 4 key directions: (1) Extending the framework to full-scene semantic segmentation of all urban elements; (2) Integrating semi-supervised learning to reduce reliance on manual annotation; (3) Exploring unsupervised model construction following semi-supervised pretraining to ultimately eliminate the need for labeled data.

  Large-Scale Airborne LiDAR Point Cloud Building Extraction Based on Improved Voxelized Deep Learning Network · 2026 · DOI

• An important direction for future work is extending the proposed framework to support multi-material and temperature-dependent elastic behavior while preserving real-time performance. Owing to the local and modular nature of Position-Based Dynamics (PBD), such extensions can be incorporated at the parameter level without altering the overall constraint structure or increasing the number of solver iterations.

  Physically plausible balloon dynamics via position-based constraints and geodesic-weighted forces · 2026 · DOI

• 3DGS models often rely on an excessive number of Gaussian primitives, resulting in redundant elements and a lack of coherent surface structure. There is a strong incentive to develop representations that retain 3DGS's photorealistic fidelity and speed while offering simpler, more structured geometric forms.

  HaloGS: Loose coupling of compact geometry and Gaussian Splats for 3D scenes · 2026 · DOI

• The paper does not fully cover adverse-weather or cross-sensor deployment. The stress test is limited to two controlled stressors: sparsification and range truncation.

  GDPNet: a hybrid GNN-Transformer with position–density-modulated attention for 3D point cloud semantic segmentation · 2026 · DOI

• We hope that providing better evaluation metrics would be an important open problem for image-to-graph generation studies. While our method achieves promising results, it is not intuitive what each metric evaluates, which has not been investigated so far.

  PlantPose: Universal Plant Skeleton Estimation via Tree-constrained Graph Generation · 2026

• Existing learning-based compression methods in the spherical domain do not exploit inter-frame correlations to reduce geometry redundancy. Traditional geometry compression methods have limitations in capturing complex motion patterns and structural dependencies.

  Inter-LPCM: Learning-based Inter-Frame Predictive Coding for LiDAR Point Cloud Compression · 2026

• Existing digital 3D models often lack structural components to function as intended. Human-annotated motions are frequently error-prone, leading to physically implausible behavior. There is a need for a novel approach to transform non-functional 3D models into functional ones.

  Functionalization via Structure Completion and Motion Rectification · 2026

• The lack of texture and color information in 3D shapes. The need for a method that can handle the cross-modality gap.

  Best Segmentation Buddies for Image-Shape Correspondence · 2026

• The study suggests that future research should focus on developing text-to-CAD systems that can support professional workflows. The study suggests that future research should investigate the use of more advanced output representations and prompt styles for text-to-CAD generation.

  Text2CAD-Bench: A Benchmark for LLM-based Text-to-Parametric CAD Generation · 2026

• The study notes that existing benchmarks focus on basic primitives and simple sketch-extrude sequences, lacking advanced features essential for real-world applications.

  Text2CAD-Bench: A Benchmark for LLM-based Text-to-Parametric CAD Generation · 2026

• Future work will focus on improving computational efficiency SIGGRAPH ’26 Conference Papers, July 19–23, 2026, Los Angeles, CA, USA.

  BrepForge: Factorized B-rep Synthesis via Wireframe Composition and Boundary-Conditioned Surface Instantiation · 2026

• Optional contour-guided densification and simplification When residual-guided latent refinement is insufficient, an optional final repair stage can insert an additional anchor into an under-parameterized segment. If the score is insufficient, build residual guidance maps and optimize only the sample-specific latent perturbation ∆z while keeping network weights frozen.

  AnchorFlow: Editable SVG Reconstruction via Sparse Anchor Point Fields · 2026

• Existing Superpoint Transformers suffer from limited performance due to low-quality superpoints. Traditional plane instance segmentation methods have limitations, such as relying on prior knowledge and manual parameter tuning.

  SPPSFormer: High-Quality Superpoint-Based Transformer for Roof Plane Instance Segmentation from Point Clouds · 2026 · DOI

• High-resolution voxels are expensive to process. Point cloud processing remains computationally intensive. Existing point cloud processing pipelines incorporate a sampling step, which can lead to loss of information.

  A Systematic Survey on Deep Learning Architectures for Point Cloud Classification and Segmentation · 2026 · DOI

• To explore the application of Inter-LPCM in various fields, such as autonomous driving and robotics.

  Inter-LPCM: Learning-based Inter-Frame Predictive Coding for LiDAR Point Cloud Compression · 2026

• To explore applications in various domains such as furniture design and robotics.

  Functionalization via Structure Completion and Motion Rectification · 2026

• Using the segment-to-segment correspondence to influence and update pretrained visual features to be correspondence-aware. Leveraging the technique to find correspondences across additional modalities, such as videos, 3D Gaussian splats, and NeRFs. Exploring image and video-driven local deformations of 3D shapes.

  Best Segmentation Buddies for Image-Shape Correspondence · 2026

• Thermal Infra-Red (TIR) images are inherently 2D and have limited spatial resolution, narrow Field of View (FoV), and poor geometric fidelity.

  Automatic generation of thermal point clouds from TIR images and LiDAR point clouds · 2026 · DOI

• Extending this work with higher-order primitives and hierarchical fitting to better capture non-planar structures. Improving the handling of semi-transparent and transparent objects. Exploring the use of HaloGS for downstream tasks like reflection modeling, collision detection, and semantic reasoning.

  HaloGS: Loose coupling of compact geometry and Gaussian Splats for 3D scenes · 2026 · DOI

• Investigating the application of GDPNet to other 3D point cloud tasks. Exploring the use of GDPNet in adverse-weather or cross-sensor deployment. Improving the efficiency of GDPNet for large-scale 3D point cloud data.

  GDPNet: a hybrid GNN-Transformer with position–density-modulated attention for 3D point cloud semantic segmentation · 2026 · DOI

• Developing more sophisticated attention mechanisms specifically tailored to the unique characteristics of point clouds. Investigating sparse attention mechanisms, point-cloud-specific tokenizations, and quantized architectures to reduce memory or compute while maintaining performance. Exploring the use of multi-modal frameworks and state-space models for point cloud processing.

  A Systematic Survey on Deep Learning Architectures for Point Cloud Classification and Segmentation · 2026 · DOI

• Improving mesh quality and robustness. Exploring applications in embodied AI, physical reasoning, and simulation.

  Fast 4D Mesh Generation by Spatio-Temporal Attention Chains · 2026

• It successfully overcomes the limitations of baseline models in handling high-frequency repetitive textures and hybrid geometric structures, effectively avoiding texture oversmoothing or structural proportion misalignment caused by single-module constraints, thereby realizing high-fidelity generation of complex hybrid topologies.

  MOC-3D: Manifold-Order Consistency for Text-to-3D Generation · 2026 · DOI

• Further improvement of the proposed method to handle extremely low point density or very large density variations. Application of the proposed method to other 3D point cloud segmentation tasks. Investigation of the use of other types of handcrafted features to improve instance segmentation accuracy.

  SPPSFormer: High-Quality Superpoint-Based Transformer for Roof Plane Instance Segmentation from Point Clouds · 2026 · DOI

• Investigation of automatic detection of regions requiring refinement. Application of the method to real-world scenarios. Further refinement of the simulation setup to better reflect real-world conditions.

  From tensor-product to truncated hierarchical B-splines: Enhancing spatial resolution in space-continuous deformation analysis based on 3D point clouds · 2026 · DOI

Most-cited papers in 3D Shape Modeling and Analysis

• PCT: Point cloud transformer · Computational Visual Media · 2021 · 1,971 citations
• Constraints on deformable models:Recovering 3D shape and nonrigid motion · Artificial Intelligence · 1988 · 544 citations
• Hierarchical geometric models for visible surface algorithms · Communications of the ACM · 1976 · 500 citations
• SuGaR: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction and High-Quality Mesh Rendering · 2024 · 451 citations
• Recovering surface shape and orientation from texture · Artificial Intelligence · 1981 · 378 citations
• Image-Based 3D Object Reconstruction: State-of-the-Art and Trends in the Deep Learning Era · IEEE Transactions on Pattern Analysis and Machine Intelligence · 2021 · 356 citations
• Recovery of the three-dimensional shape of an object from a single view · Artificial Intelligence · 1981 · 350 citations
• Point Transformer · IEEE Access · 2021 · 294 citations
• GeoTransformer: Fast and Robust Point Cloud Registration With Geometric Transformer · IEEE Transactions on Pattern Analysis and Machine Intelligence · 2023 · 272 citations
• Mixture of volumetric primitives for efficient neural rendering · ACM Transactions on Graphics · 2021 · 270 citations

Most recent work

• Periodic Vibration Gaussian: Dynamic Urban Scene Reconstruction and Real-time Rendering · International Journal of Computer Vision · 2026
• Semantic Gaussians: Open-Vocabulary Scene Understanding With 3D Gaussian Splatting · IEEE Transactions on Circuits and Systems for Video Technology · 2026
• HSBNet: Fusing semantics and anisotropic thermal diffusion fields for boundary-aware point cloud segmentation · Information Fusion · 2026
• Deep Learning-Based Point Cloud Registration: A Comprehensive Survey and Taxonomy · International Journal of Computer Vision · 2026
• LODNeuS: A Flexible Lightweight Neural Implicit Surface Representation With Unconstrained Viewpoint Rendering · IEEE Transactions on Circuits and Systems for Video Technology · 2026
• CoSurfGS: 3D Surface Gaussian Splatting with Collaborative Distributed Learning for Large-scale Scene Reconstruction · International Journal of Computer Vision · 2026
• SAGD: Boundary-Enhanced Segment Anything in 3D Gaussian via Gaussian Decomposition · IEEE Transactions on Image Processing · 2026
• 3DGeoMeshNet: A multi-scale graph auto-encoder for 3D mesh reconstruction and completion · Neurocomputing · 2026
• Weakly supervised framework for semantic segmentation of MEP point clouds · Automation in Construction · 2026
• Deep Learning-Based Joint Geometry and Attribute Up-Sampling for Large-Scale Colored Point Clouds · IEEE Transactions on Image Processing · 2026

Related topics in Engineering