Toybox

Toybox数据集由范德堡大学电气工程与计算机科学系创建，包含约230万条视频数据，主要记录了日常家用物品和玩具在第一人称视角下的手动操作和结构化变换过程，如旋转、平移和缩放。数据集通过穿戴式相机拍摄，每个对象包含12个视频片段，每个片段约20秒，展示了对象在不同变换下的连续视图。Toybox数据集旨在支持视觉对象识别及相关视觉问题的计算实验，特别是在研究人类或机器人视觉对象体验方面，为解决视觉学习和认知科学中的问题提供丰富的视觉体验数据。The Toybox Dataset was created by the Department of Electrical Engineering and Computer Science at Vanderbilt University. It comprises approximately 2.3 million video clips that primarily document first-person manual manipulation and structured transformation processes of everyday household items and toys, such as rotation, translation, and scaling. Captured via wearable cameras, each object has 12 video clips, each lasting about 20 seconds, showcasing continuous views of the object under various transformations. The Toybox Dataset aims to support computational experiments for visual object recognition and related visual research, particularly studies on human or robotic visual object perception, providing rich visual experience data to address problems in visual learning and cognitive science.

搜集汇总

背景与挑战

背景概述

Toybox数据集是一个由范德堡大学创建的大规模视频数据集，包含约230万条第一人称视角视频，记录了日常物品和玩具的手动操作与结构化变换过程（如旋转、平移和缩放），每个对象提供12个20秒的视频片段。该数据集旨在支持视觉对象识别及相关计算实验，为研究人类或机器人视觉体验以及视觉学习问题提供丰富的视觉数据资源。