Leveling3D: Leveling Up 3D Reconstruction with Feed-Forward 3D Gaussian Splatting and Geometry-Aware Generation

TL;DR

Leveling3D introduces a geometry-aware pipeline that enhances 3D reconstruction and novel-view synthesis by integrating feed-forward 3D models with geometry-consistent generation, achieving state-of-the-art results.

Contribution

It presents a novel geometry-aware leveling adapter and training strategies that improve 3D reconstruction and view synthesis, addressing missing areas in extrapolated views.

Findings

Achieves state-of-the-art performance on public datasets.

Improves extrapolated novel view quality with geometry-aware generation.

Enhances 3D reconstruction accuracy and depth estimation.

Abstract

Feed-forward 3D reconstruction has revolutionized 3D vision, providing a powerful baseline for downstream tasks such as novel-view synthesis with 3D Gaussian Splatting. Previous works explore fixing the corrupted rendering results with a diffusion model. However, they lack geometric concern and fail at filling the missing area on the extrapolated view. In this work, we introduce Leveling3D, a novel pipeline that integrates feed-forward 3D reconstruction with geometrical-consistent generation to enable holistic simultaneous reconstruction and generation. We propose a geometry-aware leveling adapter, a lightweight technique that aligns internal knowledge in the diffusion model with the geometry prior from the feed-forward model. The leveling adapter enables generation on the artifact area of the extrapolated novel views caused by underconstrained regions of the 3D representation. Specifically, to learn a more diverse distributed generation, we introduce the palette filtering strategy for training, and a test-time masking refinement to prevent messy boundaries along the fixing regions. More importantly, the enhanced extrapolated novel views from Leveling3D could be used as the inputs for feed-forward 3DGS, leveling up the 3D reconstruction. We achieve SOTA performance on public datasets, including tasks such as novel-view synthesis and depth estimation.