ThreeDWorld (TDW) dataset

This data set was created with the ThreeDWorld (TDW) software [1]. Our data set comprises twelve high quality 3D models of common household objects such as a hairbrush, a hammer or scissors. They are arranged in a4x3grid on a white floor, to simulate viewing sequences that an infant may experience while interacting with an object, such as turning an object in hand or moving relative to an object while fixating it.
Different views are created by changing viewing direction in terms of azimuth (from 0 degree to 350 degree in10 degree steps), elevation (from10 degree to 70 degree in10 degree steps), and distance (0.3 cm to 0.6 cm in 10 cm steps). This gives rise to 1,008 views per object and a total of 12,096 images, which are down-sampled to the size of 64 by 64 pixels.

The dataset was created for the paper Contrastive learning through time. In SVRHM 2021 Workshop @ NeurIPS, 2021.

[1] Chuang Gan, Jeremy Schwartz, Seth Alter, Damian Mrowca, Martin Schrimpf, James Traer, Julian DeFreitas, Jonas Kubilius, Abhishek Bhandwaldar, Nick Haber, Megumi Sano, Kuno Kim, EliasWang, Michael Lingelbach, Aidan Curtis, Kevin Tyler Feigelis, Daniel Bear, Dan Gutfreund,David Daniel Cox, Antonio Torralba, James J. DiCarlo, Joshua B. Tenenbaum, Josh Mcdermott,and Daniel LK Yamins. ThreeDWorld: A platform for interactive multi-modal physical simulation.InThirty-fifth Conference on Neural Information Processing Systems Datasets and BenchmarksTrack (Round 1), 2021. URL https://openreview.net/forum?id=db1InWAwW2T.