TL;DR
DynamicTree introduces a fast, interactive framework for realistic 4D tree animation using sparse voxel spectrum representation, enabling real-time responses and long-term motion generation for complex real trees.
Contribution
It is the first to generate long-term, interactive 3D tree motion efficiently using a novel sparse voxel spectrum representation, surpassing prior optimization-based methods.
Findings
Achieves realistic and responsive tree animations.
Outperforms existing methods in visual quality and efficiency.
Introduces the large-scale 4DTree dataset with 8,786 animated meshes.
Abstract
Generating dynamic and interactive 3D trees has wide applications in virtual reality, games, and world simulation. However, existing methods still face various challenges in generating structurally consistent and realistic 4D motion for complex real trees. In this paper, we propose DynamicTree, the first framework that can generate long-term, interactive 3D motion for 3DGS reconstructions of real trees. Unlike prior optimization-based methods, our approach generates dynamics in a fast feed-forward manner. The key success of our approach is the use of a compact sparse voxel spectrum to represent the tree movement. Given a 3D tree from Gaussian Splatting reconstruction, our pipeline first generates mesh motion using the sparse voxel spectrum and then binds Gaussians to deform the mesh. Additionally, the proposed sparse voxel spectrum can also serve as a basis for fast modal analysis under…
Peer Reviews
Decision·ICLR 2026 Conference Withdrawn Submission
1. The paper introduces a large-scale 4D tree dataset comprising thousands of animated tree meshes. 2. The authors propose a sparse voxel spectrum motion representation that supports efficient and long-term 4D generation.
1. The reviewer considers “Contribution 1” to be inadequately substantiated. The task of long-term motion generation is not new, and the framework appears to lack theoretical innovation. What specific problem does this framework address or mitigate? 2. The authors do not clarify whether the proposed 4DTree dataset will be publicly released. This raises doubts regarding the practical impact of Contribution 3. 3. While the authors propose a novel sparse voxel spectrum motion representation for eff
- A 4D tree dataset that looks realistic. - Fast simulation via modal analysis.
My concerns are centered around the quality of the results. - The animation does not seem to form a closed infinite loop. So there are some sudden changes after 3~4 seconds. - The simulation does not look realistic to me. In particular, during 1:15 - 2:20 in the supplementary video, the force response always involves trivial vibrations from other tree branches than the one branch that is directly interacted with by the external force. Looking at 2:07-2:20, both small / large forces induce trivi
- Sparse voxel spectrum representation is introduced to bridge generative modeling and physical simulation for dynamic 3D tree animation, achieving both efficient long-horizon motion generation and interactive physical response compared to previous methods. - The manuscript is well written and clearly organized, with intuitive figures and videos that effectively communicate the two-stage pipeline, spectral representation, and qualitative results. - A high-quality and relatively large-scale an
1. **Limited problem scope**. The proposed method is designed specifically for tree animation and simulation, while most compared baselines target general dynamic objects or scenes. It remains unclear whether DynamicTree can generalize beyond trees, for example, to deformable or articulated objects such as cloth, humans, or other vegetation. Clarifying whether the proposed sparse voxel spectrum framework is object-agnostic or relies on tree-specific priors would strengthen the contribution’s gen
1. The paper is well-organized and easy to read. 2. The presented method is a solid and elegant extension of Generative Image Dynamics to 3D, yielding impressive results. 3. By employs sparse voxel grids as a regular proxy, this work can robustly animate reconstructed irregular meshes and create a clean learning objective.
1. Extending generative dynamics to 3D is an interesting exploration, but it should be noted that this also sacrifices the original scalability. Unlike 2D case where training data can be collected from existing videos in a relative lower cost, its 3D data is synthesized based on prior knowledge. It constrains scaling toward more diverse and larger-scale data--despite their commendable effort, the dataset remains limited in single category, and only learning the human designed patterns somewhat u
（1）From a structural perspective, the paper is well-organized, with clear methodological exposition and illustrative diagrams, demonstrating logical coherence. （2）In terms of quality, the comprehensive experiments demonstrate that the proposed method surpasses baseline approaches in both visual quality and temporal consistency, while achieving significant speed improvements in interactive simulation. （3）The sparse voxel spectrum representation proposed in this paper presents an effective and inn
（1）The evaluation primarily focuses on swaying motions, without exploring the performance under more complex conditions such as strong wind or heavy rain effects. （2）While the method relies on synthetic training data with maximized realism in experiments, its generalization capability to diverse tree species in real-world scenarios requires further validation, and how to simplify the potential model deserves consideration. （3）Although the proposed sparse voxel spectrum representation enhances ef
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
