Fine-Detail Monocular Geometry Estimation with Self-Guided Sparse Volumetric Refinement
2026-07-20 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors explain that existing methods to estimate 3D shapes from single images tend to blur small details because they process details in flat 2D image space instead of real 3D space. To fix this, they introduce a new method called Self-Guided Sparse 3D Refinement (SSR), which lifts the 3D shape estimates into a sparse 3D grid and refines them by considering actual 3D distances. This helps the model better keep fine details like thin or small objects. Their experiments show that their approach recovers more accurate and detailed 3D geometry than previous methods.
Monocular geometry estimation3D spatial relationships2D parameterizationSparse convolutionVoxelPoint mapDepth discontinuity3D refinementMetric-scale geometry
Authors
Lingyu Kong, Ruicheng Li, Ruicheng Wang, Sicheng Xu, Chengtang Yao, Jianfeng Xiang, Jiaolong Yang
Abstract
Monocular geometry estimation has recently achieved impressive performance across diverse scenes. However, state-of-the-art models still face notable distortion in local 3D structure, especially in fine details, like thin structures and small objects. We attribute this limitation to an architectural mismatch: most current models decode 3D geometry within a 2D parameterization, where feature interactions are governed by image-plane proximity rather than true 3D spatial relationships. This inadvertently mixes features from geometrically distant surfaces, resulting in over-smoothed geometry particularly around thin or elongated structure. In this paper, we propose a fine-detail monocular geometry estimation with Self-Guided Sparse 3D Refinement (SSR) that lifts monocular geometry modeling from 2D image space to 3D space for high-fidelity metric-scale point maps. Our model lifts the coarse point map from a foundation base model onto a sparse voxel shell and refines it via SSR. The SSR employs sparse convolutions that aggregate features based on 3D spatial locality, avoiding feature mixing across depth discontinuities. Extensive experiments on diverse datasets demonstrate that our method significantly outperforms existing approaches in recovering fine detailed 3D geometry across both quantitative metrics and qualitative visualizations.