[论文] World Tracing: Generative Pixel-Aligned Geometry Beyond the Visible

论文概要

研究领域: CV 作者: Hao Zhang, Mohamed El Banani, Jen-Hao Cheng, Paul Zhang, Yi Hua, Ben Mildenhall, Christoph Lassner, Narendra Ahuja, Gengshan Yang 发布时间: 2026-06-11 arXiv: 2606.13652

中文摘要

图像到3D方法在忠实度和完整性之间权衡。我们引入World Tracing，一种生成式像素对齐几何表征，预测与观测像素对齐的3D点同时补全可见表面之外的几何。对每个像素，预测有序的3D点，第一层为可见表面，后续层表示被遮挡表面。在物体、场景和动态基准上表现强劲。

原文摘要

Image-to-3D methods trade off faithfulness and completeness. We introduce World Tracing, a generative pixel-aligned geometry representation that predicts 3D points aligned with observed pixels while completing geometry beyond visible surfaces. For each pixel, it predicts ordered 3D points where first layer is visible surface and subsequent layers represent occluded surfaces. It achieves strong performance on object, scene, and dynamic benchmarks.

#论文 #arXiv #CV #小凯