Advanced techniques using Neural Radiance Fields (NeRF), Signed Distance Fields (SDF), and Occupancy Fields have recently emerged as solutions for 3D indoor scene reconstruction. We introduce a novel two-phase learning approach, H2O-SDF, that discriminates between object and non-object regions within indoor environments. This method achieves a nuanced balance, carefully preserving the geometric integrity of room layouts while also capturing intricate surface details of specific objects. A cornerstone of our two-phase learning framework is the introduction of the Object Surface Field (OSF), a novel concept designed to mitigate the persistent vanishing gradient problem that has previously hindered the capture of high-frequency details in other methods. Our proposed approach is validated through several experiments that include ablation studies.
Our method consists of two phases: the first phase, holistic surface learning, concentrates on the global scene geometry, and the second phase, object surface learning, delves into the intricate geometrical and surface details of objects within the indoor scene. This dual-phase learning approach achieves a nuanced balance, carefully preserving the geometric integrity of room layouts while also capturing intricate surface details of specific objects.
We test our method on the ScanNet dataset and compare to state-of-the-art methods. Our method demonstrates significantly smoother reconstruction results in low-frequency areas compared to other methods, highlighting the benefits of our re-weighting scheme. Moreover, thanks to the object surface learning, H2O-SDF can better represent the fine-grained surface details of object geometries (e.g., chair legs or lamp) compared to other methods.
@misc{park2024h2osdf,
title={H2O-SDF: Two-phase Learning for 3D Indoor Reconstruction using Object Surface Fields},
author={Minyoung Park and Mirae Do and YeonJae Shin and Jaeseok Yoo and Jongkwang Hong and Joongrock Kim and Chul Lee},
year={2024},
eprint={2402.08138},
archivePrefix={arXiv},
primaryClass={cs.CV}
}