PhysicsNeRF: Physics-Guided 3D Reconstruction from Sparse Views

TL;DR

PhysicsNeRF enhances 3D reconstruction from sparse views by integrating physics-based constraints into Neural Radiance Fields, achieving superior image quality and revealing fundamental limits of sparse-view methods.

Contribution

It introduces a physics-guided framework with four constraints that significantly improves sparse-view 3D reconstruction over standard NeRFs.

Findings

Achieves 21.4 dB PSNR with only 8 views

Outperforms prior methods in sparse-view scenarios

Identifies a generalization gap of 5.7-6.2 dB in the approach

Abstract

PhysicsNeRF is a physically grounded framework for 3D reconstruction from sparse views, extending Neural Radiance Fields with four complementary constraints: depth ranking, RegNeRF-style consistency, sparsity priors, and cross-view alignment. While standard NeRFs fail under sparse supervision, PhysicsNeRF employs a compact 0.67M-parameter architecture and achieves 21.4 dB average PSNR using only 8 views, outperforming prior methods. A generalization gap of 5.7-6.2 dB is consistently observed and analyzed, revealing fundamental limitations of sparse-view reconstruction. PhysicsNeRF enables physically consistent, generalizable 3D representations for agent interaction and simulation, and clarifies the expressiveness-generalization trade-off in constrained NeRF models.