研究人员推出FoundationGeo，一个旨在改进单目度量几何估计的新型两阶段框架。该系统首先使用DINOv3和一个大型、精选的数据集学习一个仿射不变几何模型，实现了强大的跨域泛化能力。然后，它结合像素级校准场进行度量对齐和偏差校正，从而生成度量一致的三维点图。一项关键发现是相机内参覆盖范围（特别是焦距分布）对零样本泛化能力的影响，通过使用基于Blender的引擎合成数据来增强鲁棒性来解决这个问题。 AI
影响 这项研究可以改进依赖单目相机输入的应用程序中的三维重建和场景理解。
排序理由 该集群描述了一篇关于计算机视觉任务新框架的最新研究论文。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
We present FoundationGeo, a two-stage framework that explicitly bridges relative and metric prediction via spatial calibration and principled data design. Stage 1 learns a high-fidelity, affine-invariant geometry model by initializing with DINOv3 and training on a curated 10.2M-s…
arXiv:2607.11588v1 Announce Type: new Abstract: We present FoundationGeo, a two-stage framework that explicitly bridges relative and metric prediction via spatial calibration and principled data design. Stage 1 learns a high-fidelity, affine-invariant geometry model by initializi…
We present FoundationGeo, a two-stage framework that explicitly bridges relative and metric prediction via spatial calibration and principled data design. Stage 1 learns a high-fidelity, affine-invariant geometry model by initializing with DINOv3 and training on a curated 10.2M-s…