.png){kind=icon}
.png){kind=icon}
Return
Motion-driven 4D scene generation
.png){kind=icon}
PRE
AI
DOI：10.1007/s11432-024-4881-y.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Recent advances in image/video generative models and learnable dynamic scene representations have enabled the easy creation of 4D scenes. However, the question of how to achieve more controlled motion generation rather than providing high-level text prompts is not well explored. This paper presents an innovative method that leverages user-specified action paths to guide the 4D scene generation. The challenges of motion-controlled 4D scene generation mainly lie in the temporal variations of the motion and the high requirement for temporal consistency. To address these, our method incorporates an adaptive motion alignment module to enhance the naturalness of the generated motion, which dynamically synchronizes motions in the action path domain with their corresponding contents in the time domain. Additionally, to mitigate artifacts caused by excessive motion, we introduce a progressive motion drive module to ensure high-quality outputs by enhancing the control of motion. Experimental results demonstrate the superiority of our method in performing motion-controlled 4D scene generation.
Keywords:
4D generation
dynamic 3D scene
diffusion model
motion control
temporal consistency
Journal
S
IF:
7.6
Papers:
89
Citations:
0