CVRecon: Rethinking 3D Geometric Feature Learning For Neural Reconstruction

TL;DR

CVRecon introduces a novel 3D geometric feature learning framework for neural reconstruction, leveraging cost volumes and view-dependent encoding to improve reconstruction quality and detail recovery.

Contribution

The paper proposes CVRecon and RCCV, innovative methods that enhance 3D geometric feature learning by utilizing cost volumes and view-dependent information, addressing noise issues in existing approaches.

Findings

Significant improvement in reconstruction quality across metrics

Enhanced recovery of fine 3D geometric details

Robustness to occlusions and empty spaces

Abstract

Recent advances in neural reconstruction using posed image sequences have made remarkable progress. However, due to the lack of depth information, existing volumetric-based techniques simply duplicate 2D image features of the object surface along the entire camera ray. We contend this duplication introduces noise in empty and occluded spaces, posing challenges for producing high-quality 3D geometry. Drawing inspiration from traditional multi-view stereo methods, we propose an end-to-end 3D neural reconstruction framework CVRecon, designed to exploit the rich geometric embedding in the cost volumes to facilitate 3D geometric feature learning. Furthermore, we present Ray-contextual Compensated Cost Volume (RCCV), a novel 3D geometric feature representation that encodes view-dependent information with improved integrity and robustness. Through comprehensive experiments, we demonstrate that our approach significantly improves the reconstruction quality in various metrics and recovers clear fine details of the 3D geometries. Our extensive ablation studies provide insights into the development of effective 3D geometric feature learning schemes. Project page: https://cvrecon.ziyue.cool/