StayStill: a large-scale 3D idle animation dataset
Authors/Creators
-
1.
University of the Basque Country
- 2. KTH Royal Institute of Technology
-
3.
Electronic Arts (United States)
Description
Data: This repository contains the animation data presented in out paper [StayStill: a large-scale 3D idle animation dataset](https://arxiv.org/abs/2605.13693). StayStill is a dataset that contains around 650.000 frames of idle motion, encompassing general idle motion, idle motion while using a phone and idle actions. A more detailed overview of the data is explained in the [data](#data) section and in the original paper.
Related code: The code related to the dataset is available in our Github repository. This code can be used to automatically clean the data with the manual annotations and the code that we provide. It also enables to reproduce the results in the paper.
📖 Citing
If you use the StayStill dataset in any of your projects and scientific research, please cite it as follows:
bibtex@article{landa2026staystill,
author = {Atxa Landa, Eneko and Rodriguez, Igor and Lazkano, Elena and Kucherenko, Taras},
title = {StayStill: a large-scale 3D idle animation dataset},
journal = {Computer Graphics Forum},
year = {2026}
}
🧍♂️Data
Train, validation and test splits
The official train, validation and test splits, as determined in our paper, are the following:
| Split | Subjects |
| Train | 0 1 3 4 5 7 11 12 13 14 16 17 18 20 24 25 26 27 29 30 31 32 33 34 35 36 37 39 41 42 44 46 47 48 49 |
| Validation | 8 19 21 38 40 |
| test | 2 6 9 10 15 22 23 28 43 45 |
Detail
The data is divided into 2 folders:
- Freemocap: It contains the original skeleton provided by Freemocap (without the finger bones)
- Lafan: It contains the data retargeted and fitted to the LaFAN1 skeleton.
Each of these two folders has 3 subfolders:
- Idle: Contains 2 minutes long sequences of people idling
- Phone: Contains 2 minutes long sequences of people idling while using a phone
- Actions: Contains 18 different actions that are typical in idle scenarios
Each animation clip is named as so:
action_name[\_take_number]\_person_ID.bvh
The following table presents the data in a more specific manner:
| Motion Type | Action Name | Nº of frames | Duration (hh:mm:ss) | Nº of clips |
| Idle action: look up/sky | l_up | 27.943 | 15:31 | 98 |
| Idle action: look around | l_aro | 20.338 | 11:17 | 92 |
| Idle action: look down/floor | l_dow | 17.817 | 09:53 | 85 |
| Idle action: look shoes | l_sho | 15.623 | 08:40 | 75 |
| Idle action: check watch | l_wat | 11.332 | 06:17 | 92 |
| Idle action: check phone | l_pho | 21.237 | 11:47 | 89 |
| Idle action: scratch head | sc_hea | 13.199 | 07:19 | 92 |
| Idle action: scratch arm | sc_arm | 13.750 | 07:38 | 93 |
| Idle action: scratch leg | sc_leg | 11.510 | 06:23 | 77 |
| Idle action: scratch back | sc_bac | 10.900 | 06:03 | 66 |
| Idle action: touch face/chin | t_fac | 14.852 | 08:15 | 93 |
| Idle action: stretch arms | st_arm | 14.473 | 08:02 | 80 |
| Idle action: stretch back | st_bac | 12.300 | 06:50 | 71 |
| Idle action: rub eyes | sc_eye | 15.430 | 08:34 | 97 |
| Idle action: yawn | yaw | 13,766 | 07:38 | 95 |
| Idle action: look back (left) | lb_lef | 10.253 | 05:41 | 69 |
| Idle action: look back (right) | lb_rig | 10.571 | 05:52 | 75 |
| Idle action: balance l -> r | wei_lr | 11.100 | 06:10 | 48 |
| Idle action: balance r -> l | wei_rl | 9.357 | 05:11 | 47 |
| Idle actions total | -- | 275.751 | 2:33:11 | 1534 |
| General Idle | idle | 181.846 | 1:41:01 | 50 |
| Idle with a phone | phone | 187.741 | 1:44:18 | 50 |
📋 License
The provided dataset is released under the MIT License.
You are free to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the dataset, under the conditions outlined in the MIT License.
See LICENSE.txt for the full license text.