InFusionSurf: Refining Neural RGB-D Surface Reconstruction Using Per-Frame Intrinsic Refinement and TSDF Fusion Prior Learning

We introduce InFusionSurf, a novel approach to enhance the fidelity of neural radiance field (NeRF) frameworks for 3D surface reconstruction using RGB-D video frames. Building upon previous methods that have employed feature encoding to improve optimization speed, we further improve the reconstruction quality with minimal impact on optimization time by refining depth information. Our per-frame intrinsic refinement scheme addresses frame-specific blurs caused by camera motion in each depth frame. Furthermore, InFusionSurf utilizes a classical real-time 3D surface reconstruction method, the truncated signed distance field (TSDF) Fusion, as prior knowledge to pretrain the feature grid to support reconstruction details while accelerating the training. The quantitative and qualitative experiments comparing the performances of InFusionSurf against prior work indicate that our method is capable of accurately reconstructing a scene without sacrificing optimization speed. We also demonstrate the effectiveness of our per-frame intrinsic refinement and TSDF Fusion prior learning techniques via an ablation study.