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.

A Novel Plane Prior Assisted Depth Estimation of Indoor Scenes

PlaneFusion: Real-Time Indoor Scene Reconstruction With Planar Prior

PLNet: Plane and Line Priors for Unsupervised Indoor Depth Estimation

Prior surface model guided dense reconstruction via graph-cuts

HQP-MVS:High-Quality Plane Priors Assisted Multi-View Stereo for Low-Textured Areas

PlaneDepth: Self-supervised Depth Estimation via Orthogonal Planes

Neural 3D Scene Reconstruction with Indoor Planar Priors

Plane2Depth: Hierarchical Adaptive Plane Guidance for Monocular Depth Estimation

Planar Reconstruction of Indoor Scenes from Sparse Views and Relative Camera Poses

Real-time indoor scene reconstruction with Manhattan assumption

P$^2$SDF for Neural Indoor Scene Reconstruction

Indoor Scene Reconstruction From Monocular Video Combining Contextual and Geometric Priors

Efficient plane extraction using normal estimation and RANSAC from 3D point cloud

P3Depth: Monocular Depth Estimation with a Piecewise Planarity Prior

Point-Plane SLAM Using Supposed Planes for Indoor Environments

FS-Depth: Focal-and-Scale Depth Estimation from a Single Image in Unseen Indoor Scene

Depth Insight -- Contribution of Different Features to Indoor Single-image Depth Estimation

Deeper into Self-Supervised Monocular Indoor Depth Estimation

PMIndoor: Pose Rectified Network and Multiple Loss Functions for Self-Supervised Monocular Indoor Depth Estimation

AirPlanes: Accurate Plane Estimation via 3D-Consistent Embeddings

Efficient Plane-Based Optimization of Geometry and Texture for Indoor RGB-D Reconstruction