PlanarNeRF: Online Learning of Planar Primitives with Neural Radiance Fields

TL;DR

PlanarNeRF is a novel neural radiance field framework that detects dense 3D planes from visual data using online learning, combining appearance and geometry cues with a memory bank for consistent multi-frame detection.

Contribution

It introduces a new method for 3D plane detection with a neural field, a lightweight plane fitting module, and a global memory bank for cross-frame consistency, adaptable to supervised and self-supervised learning.

Findings

Effective in various scenarios

Significant improvement over existing methods

Capable of learning from sparse signals

Abstract

Identifying spatially complete planar primitives from visual data is a crucial task in computer vision. Prior methods are largely restricted to either 2D segment recovery or simplifying 3D structures, even with extensive plane annotations. We present PlanarNeRF, a novel framework capable of detecting dense 3D planes through online learning. Drawing upon the neural field representation, PlanarNeRF brings three major contributions. First, it enhances 3D plane detection with concurrent appearance and geometry knowledge. Second, a lightweight plane fitting module is proposed to estimate plane parameters. Third, a novel global memory bank structure with an update mechanism is introduced, ensuring consistent cross-frame correspondence. The flexible architecture of PlanarNeRF allows it to function in both 2D-supervised and self-supervised solutions, in each of which it can effectively learn from sparse training signals, significantly improving training efficiency. Through extensive experiments, we demonstrate the effectiveness of PlanarNeRF in various scenarios and remarkable improvement over existing works.