Trans4D: Realistic Geometry-Aware Transition for Compositional Text-to-4D Synthesis
Bohan Zeng, Ling Yang, Siyu Li, Jiaming Liu, Zixiang Zhang, Juanxi Tian, Kaixin Zhu, Yongzhen Guo, Fu-Yun Wang, Minkai Xu, Stefano Ermon, Wentao Zhang
2024-10-10
Summary
This paper presents Trans4D, a new framework that improves the generation of 4D scenes by enabling realistic transitions between complex objects and interactions.
What's the problem?
While current methods can create high-quality 4D objects or scenes, they often struggle with significant changes in shape or movement during transitions. This is especially important for applications in gaming and video production, where smooth and realistic transitions are crucial for user experience.
What's the solution?
Trans4D addresses this issue by using advanced techniques to create a detailed scene description and plan the timing of transitions. It employs multi-modal large language models (MLLMs) to generate physics-aware descriptions for initializing 4D scenes. Then, it uses a specialized geometry-aware transition network to manage how objects deform and interact during the transition. This combination allows for more accurate and visually appealing scene changes.
Why it matters?
This research is significant because it enhances how we can create dynamic and engaging visual content in 4D. By improving the quality of transitions between complex objects, Trans4D can benefit industries like gaming, animation, and virtual reality, making digital experiences more immersive and realistic.
Abstract
Recent advances in diffusion models have demonstrated exceptional capabilities in image and video generation, further improving the effectiveness of 4D synthesis. Existing 4D generation methods can generate high-quality 4D objects or scenes based on user-friendly conditions, benefiting the gaming and video industries. However, these methods struggle to synthesize significant object deformation of complex 4D transitions and interactions within scenes. To address this challenge, we propose Trans4D, a novel text-to-4D synthesis framework that enables realistic complex scene transitions. Specifically, we first use multi-modal large language models (MLLMs) to produce a physic-aware scene description for 4D scene initialization and effective transition timing planning. Then we propose a geometry-aware 4D transition network to realize a complex scene-level 4D transition based on the plan, which involves expressive geometrical object deformation. Extensive experiments demonstrate that Trans4D consistently outperforms existing state-of-the-art methods in generating 4D scenes with accurate and high-quality transitions, validating its effectiveness. Code: https://github.com/YangLing0818/Trans4D