清华大学与哈佛大学合作开发了一款名为 LangSplat 的人工智能系统，该系统在三维空间内进行高效、准确的开放式词汇搜索，为3D场景描述提供了更为精确的工具。作为首个基于3DGS（三维图形学）的3D语言场方法，LangSplat 引入了 SAM（自适应多尺度）和 CLIP（条件语言图像建模）技术，使其在开放词汇3D对象定位和语义分割任务上取得了突破性成果，性能优于现有最先进技术。
据IT之家报道，LangSplat 的显著优势在于速度和准确性。在对比测试中，该系统比 LERF（一种3D场景重建技术）快了199倍，这不仅大幅提高了工作效率，还降低了计算资源的消耗。这一突破性进展对于三维图形学、虚拟现实、增强现实以及计算机视觉等领域都具有重要意义。
英文标题：Tsinghua and Harvard Unveil LangSplat 3D Search Technology
关键词：AI system, 3D search, open vocabulary
英文关键词：AI system, 3D search, open vocabulary
News content:
In collaboration, Tsinghua University and Harvard University have developed an artificial intelligence system called LangSplat, which efficiently and accurately searches for open-ended vocabulary within three-dimensional spaces. This marks a more precise tool for describing 3D scenes. As the first method based on 3DGS (Three-Dimensional Graphics Science), LangSplat introduces SAM (Adaptive Multi-scale) and CLIP (Conditional Language-Image Modeling) technologies, achieving breakthrough results in open-vocabulary 3D object localization and semantic segmentation tasks, outperforming the most advanced existing methods.
As reported by IT Home, a significant advantage of LangSplat is its speed and accuracy. In comparative tests, the system is 199 times faster than LERF (a 3D scene reconstruction technology), not only greatly improving work efficiency but also reducing the consumption of computing resources. This groundbreaking progress is of great significance to fields such as three-dimensional graphics, virtual reality, augmented reality, and computer vision.
【来源】https://www.ithome.com/0/742/887.htm
Views: 2