ND-SDF: Learning Normal Deflection Fields for High-Fidelity Indoor Reconstruction

2024-08-22

arXiv.org

AI Summary

This work addresses the fundamental trade-off between fine geometric fidelity and global surface smoothness in neural implicit indoor reconstruction. To resolve this, we propose modeling normal deviations via a Normal Deflection Field (NDF) and introduce an unbiased ray sampling strategy guided by deflection angles to adaptively modulate geometric prior strength. Our method jointly learns signed distance functions (SDFs), normal fields, and angle-aware volumetric rendering, enabling high-fidelity recovery of complex or thin structures while significantly improving surface smoothness—especially in weakly textured regions (e.g., walls). Evaluated on multiple challenging indoor datasets, our approach achieves consistent performance gains. Crucially, it is the first to decouple dynamic normal modeling from ray sampling intensity control, establishing a novel paradigm for high-fidelity indoor reconstruction.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning with Manifolds
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Neural implicit reconstruction via volume rendering has demonstrated its effectiveness in recovering dense 3D surfaces. However, it is non-trivial to simultaneously recover meticulous geometry and preserve smoothness across regions with differing characteristics. To address this issue, previous methods typically employ geometric priors, which are often constrained by the performance of the prior models. In this paper, we propose ND-SDF, which learns a Normal Deflection field to represent the angular deviation between the scene normal and the prior normal. Unlike previous methods that uniformly apply geometric priors on all samples, introducing significant bias in accuracy, our proposed normal deflection field dynamically learns and adapts the utilization of samples based on their specific characteristics, thereby improving both the accuracy and effectiveness of the model. Our method not only obtains smooth weakly textured regions such as walls and floors but also preserves the geometric details of complex structures. In addition, we introduce a novel ray sampling strategy based on the deflection angle to facilitate the unbiased rendering process, which significantly improves the quality and accuracy of intricate surfaces, especially on thin structures. Consistent improvements on various challenging datasets demonstrate the superiority of our method.

Problem

• Recover meticulous geometry while preserving smoothness in diverse regions
• Dynamically adapt geometric priors based on sample characteristics for accuracy
• Improve reconstruction quality of intricate surfaces and thin structures

Innovation

• Learns Normal Deflection field for angular deviation
• Dynamic sample adaptation based on characteristics
• Novel ray sampling strategy for unbiased rendering