arXiv cs.AI· Zihui Gao, Ke Liu, Donny Y. Chen, Duochao Shi, Guosheng Lin, Jia-Wang Bian, Hao Chen, Chunhua Shen·· 12 小时前AI 评分34
SAGE：利用互联网视频弱监督实现 3D 几何基础模型的可扩展适配
Scalable Adaptation of 3D Geometric Foundation Models via Weak Supervision from Internet Video
AI 导读
SAGE 框架通过从原始互联网视频流中进行分层挖掘，将视频转化为训练轨迹与混合监督信号，实现 3D 几何基础模型的可扩展适配。在 MV-DUSt3R 和 VGGT 两个架构不同的骨干模型上，SAGE 在 7Scenes、TUM-RGBD、Matterport3D 等未见基准上将 Chamfer Distance 降低 20-42%，并持续提升 VGGT 表现。
正文 · 原文
Abstract:Geometric foundation models show promise in 3D reconstruction. However, their progress is severely constrained by the scarcity of diverse, large-scale 3D annotations. While static-scene Internet videos are abundant and diverse, utilizing them as a scaling source for geometric learning is challenging due to the absence of ground-truth geometry and the presence of observational noise. To address this, we propose SAGE, a framework for Scalable Adaptation of GEometric foundation models from raw video streams. SAGE leverages a hierarchical mining pipeline to transform videos into training trajectories and hybrid supervisions: (1) Informative training trajectory selection; (2) Sparse Geometric Anchoring via SfM point clouds for global structural guidance; and (3) Dense Differentiable Consistency via 3D Gaussian rendering for multi-view constraints. Besides, we also introduce a regularization strategy that replays a small fraction of supervised data to prevent catastrophic forgetting. Applied to two architecturally distinct backbones, SAGE reduces Chamfer Distance on unseen benchmarks (7Scenes, TUM-RGBD, Matterport3D) by 20-42% with MV-DUSt3R and consistently improves VGGT. SAGE establishes Internet video as a viable and scalable adaptation source for geometric foundation models. Code is available at this https URL.
| Comments: | Accepted to NeurIPS 2026 |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2602.07891 [cs.CV] |
| (or arXiv:2602.07891v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2602.07891 arXiv-issued DOI via DataCite |
Submission history
From: Zihui Gao [view email]
[v1]
Sun, 8 Feb 2026 09:53:21 UTC (7,420 KB)
[v2]
Fri, 16 Oct 2026 12:13:07 UTC (7,424 KB)
来源：arXiv cs.AI · arxiv.org