OCTScenes

OCTScenes是由复旦大学计算机科学与技术学院创建的一个多功能真实世界桌面场景数据集，专为以物体为中心的学习设计。该数据集包含5000个桌面场景，每个场景包含15种不同物体，共计60帧，覆盖360度视角。OCTScenes旨在作为比较、评估和分析以物体为中心的学习方法的基准。数据集通过配备3D相机的机器人进行采集，能够同时满足基于单图像、视频和多视角的以物体为中心的学习方法的评估需求。此外，OCTScenes还提供了分割基础真值，用于评估。该数据集适用于推动现有方法的发展，并激发对真实世界场景的适应性研究。

OCTScenes is a versatile real-world desktop scene dataset created by the School of Computer Science, Fudan University, designed specifically for object-centric learning. This dataset contains 5000 desktop scenes, each with 15 distinct objects and 60 frames covering 360-degree viewpoints. OCTScenes aims to serve as a benchmark for comparing, evaluating, and analyzing object-centric learning approaches. The dataset is collected via a robot equipped with a 3D camera, which can meet the evaluation requirements of single-image, video-based, and multi-view-based object-centric learning methods simultaneously. In addition, OCTScenes also provides segmentation ground truth for evaluation. This dataset is suitable for advancing the development of existing methods and inspiring research on adaptation to real-world scenes.