![[2602.01844] CloDS: Visual-Only Unsupervised Cloth Dynamics Learning in Unknown Conditions](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.01844] CloDS: Visual-Only Unsupervised Cloth Dynamics Learning in Unknown Conditions
Summary
The paper presents CloDS, an unsupervised learning framework for cloth dynamics using visual data, addressing limitations of existing methods that require known physical properties.
Why It Matters
CloDS advances the field of computer vision by enabling the simulation of cloth dynamics without prior knowledge of physical properties, which is crucial for applications in robotics, animation, and virtual reality. This approach enhances the generalization capabilities of models in unknown conditions, making it a significant step forward in dynamic system modeling.
Key Takeaways
- CloDS introduces a novel unsupervised learning framework for cloth dynamics.
- The method utilizes multi-view visual observations to learn dynamics without prior physical property knowledge.
- A dual-position opacity modulation technique is employed to handle complex deformations.
- The framework demonstrates strong generalization capabilities for unseen configurations.
- Code and visualization results are made publicly available for further research.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.01844 (cs) [Submitted on 2 Feb 2026 (v1), last revised 20 Feb 2026 (this version, v2)] Title:CloDS: Visual-Only Unsupervised Cloth Dynamics Learning in Unknown Conditions Authors:Yuliang Zhan, Jian Li, Wenbing Huang, Wenbing Huang, Yang Liu, Hao Sun View a PDF of the paper titled CloDS: Visual-Only Unsupervised Cloth Dynamics Learning in Unknown Conditions, by Yuliang Zhan and 5 other authors View PDF HTML (experimental) Abstract:Deep learning has demonstrated remarkable capabilities in simulating complex dynamic systems. However, existing methods require known physical properties as supervision or inputs, limiting their applicability under unknown conditions. To explore this challenge, we introduce Cloth Dynamics Grounding (CDG), a novel scenario for unsupervised learning of cloth dynamics from multi-view visual observations. We further propose Cloth Dynamics Splatting (CloDS), an unsupervised dynamic learning framework designed for CDG. CloDS adopts a three-stage pipeline that first performs video-to-geometry grounding and then trains a dynamics model on the grounded meshes. To cope with large non-linear deformations and severe self-occlusions during grounding, we introduce a dual-position opacity modulation that supports bidirectional mapping between 2D observations and 3D geometry via mesh-based Gaussian splatting in video-to-geometry grounding stage. It jointly considers the absolute and relative po...