Introduction

Toybox is designed to enable an improved understanding of small sample learning and hand object scene interaction. This dataset was developed by a group of researchers from AIVAS Lab, Vanderbilt University.


About the Dataset

Toybox contains 12 categories, roughly grouped into three super-categories:

To maximize the usefulness of Toybox for comparisons with studies of human learning, all 12 of these categories are among the most common early-learned nouns for typically developing children in the U.S. (http://wordbank.stanford.edu/)

All videos were recorded using Pivothead Original Series wearable cameras, which are worn like a pair of sunglasses and have the camera located just above the bridge of the wearer’s nose. Specific settings of the camera are shown below:


Download

The dataset can be downloaded from Zenodo and is split into three super-categories. The entire dataset is 110GB. Each chunk of the dataset can be downloaded as a TAR archive that contains videos for the four object categories in a given super-category.

For each of the four object categories, the archive contains 30 folders, each corresponding to an individual object in that category, labeled by the name and ID of the object, for example: “airplane_01”, “airplane_02”, etc. Each labeled folder contains 12 video files in MP4 format, for the 12 different video transformations of the object.


Publication

If you wish to cite this work, please use the citation below:

@misc{1806.06034,
Author = {Xiaohan Wang and Tengyu Ma and James Ainooson and Seunghwan Cha and Xiaotian Wang and Azhar Molla and Maithilee Kunda},
Title = {Seeing Neural Networks Through a Box of Toys: The Toybox Dataset of Visual Object Transformations},
Year = {2018},
Eprint = {arXiv:1806.06034},
}

@InProceedings{Wang_2017_ICCV,
author = {Wang, Xiaohan and Eliott, Fernanda M. and Ainooson, James and Palmer, Joshua H. and Kunda, Maithilee},
title = {An Object Is Worth Six Thousand Pictures: The Egocentric, Manual, Multi-Image (EMMI) Dataset},
booktitle = {The IEEE International Conference on Computer Vision (ICCV) Workshops},
month = {Oct},
year = {2017}
}


Team


Core Team Members (alphabetical)
ainooson
James Ainooson
cha
Seunghwan Cha
kunda
Maithilee Kunda


fernanda
Fernanda M. Eliott
josh
Joshua Palmer
hwang
Tengyu Ma

hwang
Azhar Molla
hwang
Xiaohan Wang
hwang
Xiaotian Wang

Other Contributors (alphabetical)
ellis
Ellis Brown
aneesha
Aneesha Dasari
max
Max Degroot


joe
Joseph Eilbert
joel
Joel M. Michelson
soobeen
Soobeen Park
harsha
Harsha P. Vankayalapati

Acknowledgement

Thanks also to Linda Smith, Chen Yu, Fuxin Li, and Jim Rehg for early discussions influencing this research. This research was funded in part by a Vanderbilt Discovery Grant, “New explorations in visual object recognition,” and by NSF award #1730044.