Overview
4DHumanOutfit is a new dataset of 4D human motion sequences, sampled densely in space and time, with 20 actors, dressed in 7 outfits each, and performing 11 motions exhibiting large displacements in each outfit. We designed 4DHumanOutfit to enable the combined analysis of shape, outfit and motions with humans. This results in a dataset shaped as a cube of data containing 4D motion sequences with three different factors that vary along the axes identity, outfit, and motion.
This dataset, which was acquired in Kinovis (total capture time was 5206 hours, over more than 6 months) allows to study dressed humans in motion, and to compare different algorithms on a same dataset, which we hope will serve as a strong benchmark for the computer vision research community.
More details on the dataset can be found in the article:
To appear in Computer Vision and Image Understanding.
Bibtex information
@article{4dhumanoutfit,
author = "Matthieu Armando and Laurence Boissieux and Edmond Boyer and Jean-Sebastien Franco and Martin Humenberger and Christophe Legras and Vincent Leroy and Mathieu Marsot and Julien Pansiot and Sergi Pujades and Rim Rekik and Gregory Rogez and Anilkumar Swamy and Stefanie Wuhrer",
title = "4DHumanOutfit: a multi-subject 4D dataset of human motion sequences in varying outfits exhibiting large displacements",
journal = "Computer Vision and Image Understanding",
year = to appear
}
| This work was done in collaboration with |  . |
Identities
- 10 females
- 10 males
1 male and 1 female actor will not be released to allow for future evaluations on unobserved data.
Outfits
Female
- own: the actor’s own arbitrary clothes on arrival in the studio, this outfit was therefore unique to each actor, with the purpose to increase variability
- tig: socks, dotted white leggings, dotted salmon tank top, pink swimming cap (minimal clothing)
- sho: white and pink sneakers, yellow shorts, purple T-shirt
- jea: yellow ballerinas, jeans, green and pink flowery shirt
- cos: yellow ballerinas, jeans, flowery purple dress
- opt: pink flip-flops or cream high heels, short grey dress or long loose blue dress or long tight red dress; as apparel for females offers more diversity than for males, we chose 3 optional outfits, using different materials and shapes to increase variability
- hidden: we recorded one additional outfit (white and pink sneakers, jeans, lavender hoodie) which will not be released to allow for future evaluations on unobserved data
Male
- own: the actor own arbitrary clothes on arrival in the studio
- tig: socks, beige shorts, grey tank top, blue swimming cap (minimal clothing)
- sho: blue and white sneakers, beige shorts, orange T-shirt with picture
- jea: black moccasins, jeans, grey and white striped shirt
- cos: black moccasins, dark costume trousers, grey and white striped shirt, dark costume jacket
- pt: black moccasins, dark costume trousers, grey and white striped shirt, beige trench coat
- hidden: we recorded one additional outfit (blue and white sneakers, jeans, blue hoodie) which will not be released to allow for future evaluations on unobserved data
Motions
- walk: a simple walk across the studio
- avoid: a walk with last-second obstacle avoidance
- back: a walk with a U-turn
- torso: a walk with a torso rotation to look backwards
- run: a jog / run across the studio
- jump: jump on the spot
- dance: a dance with both legs and arms wide motion
- hop: hopscotch
- 2 free motions: to be chosen by the actor to increase the variability of the dataset, this included mostly martial art, dance and other sport motions
- hidden: we recorded one additional motion (a walk with an arm wave) which will not be released to allow for future evaluations on unobserved data
Download
The data is released for non-commercial research purposes. The release is subject to a license that includes compliance with the General Data Protection Regulation (GDPR) of the European Union, and requests need to be signed by the legal representatives of the interested institutions.
The 4D Human Outfit Catalogue is available for consultation, the data dowload requires an authentification.
To obtain this authentification, please contact us and provide information on the name and address of your research institution, the name, title and e-mail address of the legal representative of your institution. If your institution is within the European Union, we further require the e-mail address and postal address of the Data Protection Officer (DPO) of your institution.
