NeuDonatello: Uncertainty-Aware SDF Learning for High-Fidelity Neural Surface Reconstruction
AAAI'26 (Under Review)
Abstract
Neural surface reconstruction has emerged as a powerful paradigm for recovering high-quality 3D surfaces from multi-view images. However, recovering accurate geometry solely from RGB images remains challenging due to uncertainties arising from textureless regions, occlusions, and inherent scene ambiguities. Existing methods often overlook such uncertainties, leading to inaccurate estimates of the signed distance function (SDF).
We introduce NeuDonatello, a novel framework that models and leverages SDF uncertainty to improve surface reconstruction. Central to our approach is to model spatially varying uncertainty using a Monte Carlo sampling strategy. Using this uncertainty, we develop an adaptive regularization that selectively strengthens geometric constraints where RGB supervision is unreliable, avoiding incorrect surface reconstruction. We further introduce an uncertainty-aware scale parameter for the SDF-to-density conversion. Conditioned on uncertainty, this design enables more accurate modeling of spatially varying densities. Extensive experiments demonstrate that NeuDonatello achieves state-of-the-art reconstruction accuracy, with robust performance across diverse scenes using only posed RGB images.
Method
- Uncertainty Modeling: NeuDonatello explicitly model SDF uncertainty to distinguish between confident and uncertain regions in the learned representation. We propagate the uncertainty through Monte Carlo sampling and supervise using NLL loss, effectively capturing local geometric uncertainty from posed multi-view RGB images.
- Uncertainty-Aware Adaptive Geometric Regularization: We develop an adaptive regularization that selectively strengthens geometric constraints where RGB supervision is unreliable, avoiding incorrect surface reconstruction.
- Uncertainty-Aware SDF-to-Density Conversion: We introduce an uncertainty-aware scale parameter for the SDF-to-density conversion. Conditioned on uncertainty, this design enables more accurate modeling of spatially varying densities and reduces density bias.
- Extensive Experiments: We conduct extensive experiments on the ScanNet++ dataset and Tanks and Temples dataset, demonstrating the effectiveness of our approach.
Citation
@unpublished{choi2025neudonatello,
title = {NeuDonatello: Uncertainty-Aware SDF Learning for High-Fidelity Neural Surface Reconstruction},
author = {Choi, Alvin Jinsung and Kim, Wanhee and Kim, Taeyun and Hong, Dasol and Lee, Wooju and Myung, Hyun},
note = {Under review at the AAAI Conference on Artificial Intelligence (AAAI), 2026},
year = {2025}
}
Relevant Projects
powered by Academic Project Page Template