Abstract:Given only a set of images, neural implicit surface representation has shown its capability in 3D surface reconstruction. However, as the nature of per-scene optimization is based on the volumetric rendering of color, previous neural implicit surface reconstruction methods usually fail in low-textured regions, including the floors, walls, etc., which commonly exist for indoor scenes. Being aware of the fact that these low-textured regions usually correspond to planes, without introducing additional ground-truth supervisory signals or making additional assumptions about the room layout, we propose to leverage a novel Pseudo Plane-regularized Signed Distance Field (P$^2$SDF) for indoor scene reconstruction. Specifically, we consider adjacent pixels with similar colors to be on the same pseudo planes. The plane parameters are then estimated on the fly during training by an efficient and effective two-step scheme. Then the signed distances of the points on the planes are regularized by the estimated plane parameters in the training phase. As the unsupervised plane segments are usually noisy and inaccurate, we propose to assign different weights to the sampled points on the plane in plane estimation as well as the regularization loss. The weights come by fusing the plane segments from different views. As the sampled rays in the planar regions are redundant, leading to inefficient training, we further propose a keypoint-guided rays sampling strategy that attends to the informative textured regions with large color variations, and the implicit network gets a better reconstruction, compared with the original uniform ray sampling strategy. Experiments show that our P$^2$SDF achieves competitive reconstruction performance in Manhattan scenes. Further, as we do not introduce any additional room layout assumption, our P$^2$SDF generalizes well to the reconstruction of non-Manhattan scenes.

SDFMAP: Neural Signed Distance Fields for Mapping and Positioning in Real-Time

HI-SLAM: Monocular Real-time Dense Mapping with Hybrid Implicit Fields

NeSLAM: Neural Implicit Mapping and Self-Supervised Feature Tracking With Depth Completion and Denoising

I2-SDF: Intrinsic Indoor Scene Reconstruction and Editing Via Raytracing in Neural SDFs

Depth Reconstruction with Neural Signed Distance Fields in Structured Light Systems

iDF-SLAM: End-to-End RGB-D SLAM with Neural Implicit Mapping and Deep Feature Tracking

N$^{3}$-Mapping: Normal Guided Neural Non-Projective Signed Distance Fields for Large-scale 3D Mapping

NC-SDF: Enhancing Indoor Scene Reconstruction Using Neural SDFs with View-Dependent Normal Compensation

Neural Implicit Mapping via Nested Neighborhoods

TivNe-SLAM: Dynamic Mapping and Tracking via Time-Varying Neural Radiance Fields

Fast Learning of Signed Distance Functions from Noisy Point Clouds via Noise to Noise Mapping

Real-Time Dense Visual SLAM with Neural Factor Representation

3D LiDAR Mapping in Dynamic Environments Using a 4D Implicit Neural Representation

NeUDF: Leaning Neural Unsigned Distance Fields with Volume Rendering

P$^2$SDF for Neural Indoor Scene Reconstruction

VI-NeRF-SLAM: a real-time visual–inertial SLAM with NeRF mapping

Fast Monocular Scene Reconstruction with Global-Sparse Local-Dense Grids

DDN-SLAM: Real-time Dense Dynamic Neural Implicit SLAM

Structerf-SLAM: Neural implicit representation SLAM for structural environments

Neural Implicit Representation for Highly Dynamic LiDAR Mapping and Odometry

ENeRF-SLAM:A Dense Endoscopic SLAM with Neural Implicit Representation