NeRF Papers for Thu May 2 2024
Use the forward/backward buttons at the top left to move between days. Use the home button to jump to latest day. Use the arrow keys on the keyboard or in the bottom right to move between categories and papers. ESC enters an overview mode. On mobile touch inputs with swipes are enabled.Priors and Generative
Priors can either aid in the reconstruction or can be used in a generative manner. For example, in the reconstruction, priors either increase the quality of neural view synthesis or enable reconstructions from sparse image collections.NC-SDF: Enhancing Indoor Scene Reconstruction Using Neural SDFs with View-Dependent Normal Compensation
Ziyi Chen, Xiaolong Wu, Yu Zhang
Figure 1. Comparison between baseline and NC-SDF. Stateof-the-art neural implicit surface representations produce suboptimal reconstructions with noisy or missing surfaces, primarily due to multi-view inconsistency between monocular geometric priors. Our NC-SDF introduces a view-dependent normal compensation model to adaptively learn and correct biases in normal priors. This approach enables the recovery of intricate geometric details while ensuring smoothness in texture-less areas within reconstructions.
NC-SDF is a neural signed distance field (SDF) 3D reconstruction framework that addresses multi-view inconsistency issues by integrating view-dependent biases into the normal priors. This correction method enhances reconstruction quality by improving global consistency and local details. Additionally, NC-SDF introduces informative pixel sampling and a hybrid geometry modeling approach, showing superior performance compared to existing methods in synthetic and real-world datasets.
Fundamentals
These papers address more fundamental problems of view-synthesis with NeRF methods.NeRF-Guided Unsupervised Learning of RGB-D Registration
Zhinan Yu, Zheng Qin, Yijie Tang, Yongjun Wang, Renjiao Yi, Chenyang Zhu, Kai Xu
Fig. 1: We propose NeRF-UR, a frame-to-model optimization framework for unsupervised RGB-D registration. The registration is first bootstrapped with synthetic data, and then fine-tuned on real-world data under guidance of NeRF (top). Although the frame-to-frame method can successfully register the easy case (bottom-left), it cannot register the case with lighting changes and low overlap (bottom-right). On the contrary, our method effectively register the hard case.
NeRF-UR introduces a frame-to-model optimization approach for unsupervised RGB-D registration, using a neural radiance field (NeRF) as a global scene model. By leveraging the consistency between input frames and NeRF-rerendered frames, the method improves robustness in scenarios with poor multi-view consistency. The framework includes a synthetic dataset, Sim-RGBD, for initial model training before unsupervised fine-tuning on real data, achieving superior performance on ScanNet and 3DMatch datasets.