Gen4D: Synthesizing Humans and Scenes in the Wild
By: Jerrin Bright , Zhibo Wang , Yuhao Chen and more
Potential Business Impact:
Creates realistic sports videos for AI training.
Lack of input data for in-the-wild activities often results in low performance across various computer vision tasks. This challenge is particularly pronounced in uncommon human-centric domains like sports, where real-world data collection is complex and impractical. While synthetic datasets offer a promising alternative, existing approaches typically suffer from limited diversity in human appearance, motion, and scene composition due to their reliance on rigid asset libraries and hand-crafted rendering pipelines. To address this, we introduce Gen4D, a fully automated pipeline for generating diverse and photorealistic 4D human animations. Gen4D integrates expert-driven motion encoding, prompt-guided avatar generation using diffusion-based Gaussian splatting, and human-aware background synthesis to produce highly varied and lifelike human sequences. Based on Gen4D, we present SportPAL, a large-scale synthetic dataset spanning three sports: baseball, icehockey, and soccer. Together, Gen4D and SportPAL provide a scalable foundation for constructing synthetic datasets tailored to in-the-wild human-centric vision tasks, with no need for manual 3D modeling or scene design.
Similar Papers
Avatar4D: Synthesizing Domain-Specific 4D Humans for Real-World Pose Estimation
CV and Pattern Recognition
Creates realistic people for sports training videos.
Advances in 4D Generation: A Survey
CV and Pattern Recognition
Creates moving 3D worlds from your ideas.
Video4DGen: Enhancing Video and 4D Generation through Mutual Optimization
Graphics
Creates realistic moving 3D objects from videos.