Syn4D: A Multiview Synthetic 4D Dataset
Zeren Jiang, Yushi Lan, Yihang Luo, Yufan Deng, Zihang Lai + 6 more
TLDR
Syn4D is a new multiview synthetic 4D dataset with ground-truth annotations for dynamic scenes, aiding 3D reconstruction and tracking.
Key contributions
- Introduces Syn4D, a multiview synthetic dataset for dynamic scenes.
- Provides ground-truth camera motion, depth maps, dense tracking, and human pose annotations.
- Enables unprojection of any pixel into 3D at any time and from any camera.
- Evaluated for 4D scene reconstruction, 3D point tracking, and human pose estimation.
Why it matters
The scarcity of high-quality datasets has hindered progress in dynamic 3D scene reconstruction and tracking. Syn4D provides crucial ground-truth data, directly addressing this limitation and accelerating research in spatiotemporal modeling.
Original Abstract
Dense 3D reconstruction and tracking of dynamic scenes from monocular video remains an important open challenge in computer vision. Progress in this area has been constrained by the scarcity of high-quality datasets with dense, complete, and accurate geometric annotations. To address this limitation, we introduce Syn4D, a multiview synthetic dataset of dynamic scenes that includes ground-truth camera motion, depth maps, dense tracking, and parametric human pose annotations. A key feature of Syn4D is the ability to unproject any pixel into 3D to any time and to any camera. We conduct extensive evaluations across multiple downstream tasks to demonstrate the utility and effectiveness of the proposed dataset, including 4D scene reconstruction, 3D point tracking, geometry-aware camera retargeting, and human pose estimation. The experimental results highlight Syn4D's potential to facilitate research in dynamic scene understanding and spatiotemporal modeling.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.