|
--- |
|
license: other |
|
license_name: s-lab-license |
|
license_link: LICENSE |
|
language: |
|
- en |
|
tags: |
|
- Human Pose and Shape Estimation |
|
- 3D Human Recovery |
|
- Parametric Humans |
|
- Synthetic Data |
|
pretty_name: GTA-Human |
|
size_categories: |
|
- 100B<n<1T |
|
--- |
|
|
|
Please visit our [Homepage](https://caizhongang.com/projects/GTA-Human/) and [Toolbox](https://github.com/caizhongang/gta-human_toolbox/tree/main) for more information. |
|
|
|
Abstract: |
|
Image- and video-based 3D human recovery ( i.e. , pose and shape estimation) have achieved substantial progress. |
|
However, due to the prohibitive cost of motion capture, existing datasets are often limited in scale and diversity. |
|
In this work, we obtain massive human sequences by playing the video game with automatically annotated 3D ground truths. |
|
Specifically, we contribute GTA-Human, a large-scale 3D human dataset generated with the GTA-V game engine, featuring |
|
a highly diverse set of subjects, actions, and scenarios. More importantly, we study the use of game-playing data |
|
and obtain five major insights. First , game-playing data is surprisingly effective. A simple frame-based baseline |
|
trained on GTA-Human outperforms more sophisticated methods by a large margin. For videobased methods, GTA-Human |
|
is even on par with the in-domain training set. Second , we discover that synthetic data provides critical complements |
|
to the real data that is typically collected indoor. We highlight that our investigation into domain gap provides |
|
explanations for our data mixture strategies that are simple yet useful, which offers new insights to the research |
|
community. Third , the scale of the dataset matters. The performance boost is closely related to the additional |
|
data available. A systematic study on multiple key factors (such as camera angle and body pose) reveals that the |
|
model performance is sensitive to data density. Fourth , the effectiveness of GTA-Human is also attributed to the |
|
rich collection of strong supervision labels (SMPL parameters), which are otherwise expensive to acquire in real |
|
datasets. Fifth , the benefits of synthetic data extend to larger models such as deeper convolutional neural networks |
|
(CNNs) and Transformers, for which a significant impact is also observed. We hope our work could pave the way for |
|
scaling up 3D human recovery to the real world. |
|
|
|
|
