TO-Scene

我们介绍了一个以桌面场景为重点的大规模数据集-场景，其中包含三种变体的20,740场景。为了获取数据，我们设计了一个高效且可扩展的框架，其中开发了众包UI，以将CAD对象从ModelNet和ShapeNet转移到ScanNet的表格上，然后将输出的桌面场景模拟为真实扫描并自动注释。

We introduce a large-scale dataset focused on desktop scenarios, named Scene, which contains 20,740 scenes with three variants each. To collect the dataset, we designed an efficient and scalable framework. A crowdsourcing UI was developed within this framework to transfer CAD objects from ModelNet and ShapeNet onto the tables in ScanNet scenes. Subsequently, the generated desktop scenes were simulated to match real-world scans and automatically annotated.

搜集汇总

背景与挑战

背景概述

TO-Scene是一个专注于桌面场景的大规模数据集，包含20,740个场景的三种变体。该数据集通过高效可扩展的框架构建，将CAD对象从现有数据集转移到模拟真实扫描的桌面场景中，并自动生成注释。