SurGe: Improved Surface Geometry in Point Maps
2026-05-29 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors point out that current 3D reconstruction methods can capture the overall shape well but often miss small details on surfaces. To better measure these local mistakes, they created a new way to check how the tiny surface angles match up. They also designed two new techniques to improve detail accuracy: one that helps the model learn better local 3D changes, and another that uses a special attention method to refine features step-by-step. Their approach, called SurGe, performs better on standard tests, especially in capturing local surface details.
3D reconstructionpoint maplocal surface geometrynormal metricdepth-normalized finite differencesNeighborhood Attentiondecodermonocular geometryAbsRelfeature upsampling
Authors
Karim Knaebel, Gonzalo Martin Garcia, Christian Schmidt, Ilya Fradlin, Lucas Nunes, Daan de Geus, Bastian Leibe
Abstract
Recent feedforward 3D reconstruction methods predict point maps and estimate global 3D geometry remarkably well. However, their predictions still exhibit inaccurate local surface geometry, which is clearly visible qualitatively but only weakly reflected in common metrics. To make these errors more explicit in evaluation, we introduce a point map normal metric that evaluates the local surface orientation induced by neighboring 3D predictions. To reduce these errors, we propose two complementary components: a point gradient matching loss that supervises depth-normalized 3D finite differences, and a Neighborhood Attention Decoder (NAD) that progressively upsamples features and uses Neighborhood Attention for local feature mixing. Across eight zero-shot monocular geometry benchmarks, our model, SurGe, achieves the best average rank for global point map AbsRel and consistently improves local point map and point map normal evaluations.