Given a set of a reference RGBD views of an indoor environment, and a new viewpoint, our goal is to predict the view from that location.

Neural Information Processing Systems http://nips.cc/

The process of 3D scene reconstruction can be affected by numerous uncertainty sources in real-world scenes. While Neural Radiance Fields (NeRFs) and 3D Gaussian Splatting (GS) achieve high-fidelity rendering, they lack built-in mechanisms to directly address or quantify uncertainties arising from the presence of noise, occlusions, confounding outliers, and imprecise camera pose inputs. In this paper, we introduce a taxonomy that categorizes different sources of uncertainty inherent in these methods. Moreover, we extend NeRF- and GS-based methods with uncertainty estimation techniques, including learning uncertainty outputs and ensembles, and perform an empirical study to assess their ability to capture the sensitivity of the reconstruction. Our study highlights the need for addressing various uncertainty aspects when designing NeRF/GS-based methods for uncertainty-aware 3D reconstruction.

Abstract-- Autonomous robotic tasks require actively perceiving the environment to achieve application-specific goals. By incrementally adding new measurements into our image collection, our approach efficiently explores an unknown scene in a mapless manner. Our planning experiments using synthetic and real-world data verify that our uncertainty-guided Figure 1: Our novel NBV planning framework exploits uncertainty approach finds informative images leading to more accurate estimation in image-based neural rendering to guide measurement scene representations when compared against baselines. Brighter frustums indicate higher average uncertainty from the view. While In this work, we present a new framework for iteratively showing promising results, these studies follow an active planning the next best view (NBV) for an RGB camera to learning [15] paradigm to collect the most informative, i.e. explore an unknown scene. Given a limited measurement most uncertain, images for periodically re-training a NeRF budget, our goal is to actively position the sensor to gather to improve the scene representation with minimal data.

In natural contexts, visual object recognition in humans is remarkably fast, reliable, and viewpoint invariant. The present approach to object recognition is "view-based" (e.g.

In natural contexts, visual object recognition in humans is remarkably fast, reliable, and viewpoint invariant. The present approach to object recognition is "view-based" (e.g.

In natural contexts, visual object recognition in humans is remarkably fast, reliable, and viewpoint invariant. The present approach to object recognition is "view-based" (e.g.