Details Enhancement in Unsigned Distance Field Learning for High-fidelity 3D Surface Reconstruction

2024-06-01

arXiv.org

AI Summary

UDFs face fundamental challenges in reconstructing open surfaces and complex internal structures, including non-differentiability of the zero-level set, gradient vanishing, proliferation of local minima, and gradient oscillations. To address these, we propose a detail-enhanced UDF framework: (1) integrating normal alignment with SIREN-based neural networks to improve geometric detail modeling; (2) introducing an adaptive weighted Eikonal constraint to mitigate gradient degradation; (3) adopting an unconditional MLP representation to relax the non-negativity constraint, enabling robust modeling of open surfaces; and (4) proposing the DCUDF algorithm for stable extraction of the locally minimal average-distance surface from the UDF. Our method enables stable training directly from unoriented point clouds, significantly improving reconstruction accuracy and surface fidelity. Extensive experiments demonstrate consistent state-of-the-art performance across multiple benchmarks, outperforming existing UDF-based approaches.

Technology Category

• Machine Learning: Learning with Manifolds
• Computer Vision: 3D Computer Vision
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

While Signed Distance Fields (SDF) are well-established for modeling watertight surfaces, Unsigned Distance Fields (UDF) broaden the scope to include open surfaces and models with complex inner structures. Despite their flexibility, UDFs encounter significant challenges in high-fidelity 3D reconstruction, such as non-differentiability at the zero level set, difficulty in achieving the exact zero value, numerous local minima, vanishing gradients, and oscillating gradient directions near the zero level set. To address these challenges, we propose Details Enhanced UDF (DEUDF) learning that integrates normal alignment and the SIREN network for capturing fine geometric details, adaptively weighted Eikonal constraints to address vanishing gradients near the target surface, unconditioned MLP-based UDF representation to relax non-negativity constraints, and DCUDF for extracting the local minimal average distance surface. These strategies collectively stabilize the learning process from unoriented point clouds and enhance the accuracy of UDFs. Our computational results demonstrate that DEUDF outperforms existing UDF learning methods in both accuracy and the quality of reconstructed surfaces.

Problem

• Challenges in high-fidelity 3D reconstruction using UDFs
• Non-differentiability and vanishing gradients in UDF learning
• Enhancing accuracy and detail in 3D surface reconstruction

Innovation

• Integrates normal alignment and SIREN network
• Uses adaptively weighted Eikonal constraints
• Employs unconditioned MLP-based UDF representation