--- license: mit language: - en size_categories: - n<1K --- ## Overview [MOSIV](https://arxiv.org/abs/2603.06022) is a synthetic dataset designed for the task of **multi-object system identification from videos**, where the goal is to recover both the **4D geometry (time-varying 3D shape)** and the **physical properties** (e.g., stiffness, friction, plasticity) of multiple interacting objects directly from multi-view video observations. ![MOSIV Teaser](assets/3obj_data.png) MOSIV features **contact-rich, multi-object interactions** with diverse materials and complex dynamics. Each sequence contains multiple objects undergoing collisions, deformation, and motion, along with ground-truth physical parameters and simulation data. The dataset is generated using [Genesis simulator](https://github.com/Genesis-Embodied-AI/Genesis) and supports tasks such as: - Dynamic 3D / 4D reconstruction - Physics-based system identification - Multi-object interaction modeling - Future state prediction and simulation ## Dataset Structure The dataset is organized hierarchically based on the number of objects and sequence IDs. ``` MOSIV/ ├── 2objs/ │ ├── 01/ │ │ ├── data/ │ │ ├── masks/ │ │ ├── masks_vis/ │ │ ├── point_clouds/ │ │ ├── videos/ │ │ ├── all_data.json │ │ ├── metadata.json │ │ ├── points3d.ply │ │ ├── transforms_train.json │ │ ├── transforms_val.json │ │ └── transforms_test.json │ ├── 02/ │ └── … ├── 3objs/ │ ├── 012/ │ ├── 013/ │ └── … ``` Citing MOSIV If you find this dataset useful in your work, please consider citing our paper: ``` @misc{liu2026mosivmultiobjectidentificationvideos, title={MOSIV: Multi-Object System Identification from Videos}, author={Chunjiang Liu and Xiaoyuan Wang and Qingran Lin and Albert Xiao and Haoyu Chen and Shizheng Wen and Hao Zhang and Lu Qi and Ming-Hsuan Yang and Laszlo A. Jeni and Min Xu and Yizhou Zhao}, year={2026}, eprint={2603.06022}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2603.06022}, } ```