**DynVideo-E: Innovative Approach to Human-Centric Video Editing**

This poster from the Applied Research Center (ARC) at the National University of Singapore (NUS) presents a cutting-edge technology named DynVideo-E, which harnesses Dynamic NeRF for large-scale video editing. The research team, including Jia-Wei Liu, Yan-Pei Cao, Jay Zhangjie Wu, and Weijia Mao, explores advanced 3D representation for challenging human-centric video editing scenarios. The main highlight is the development of a video-3D representation system that maintains long-term consistency, even with significant motion or viewpoint changes.

Key features and findings:
- A system to generate dynamic human editing designs using image-based 3D representations.
- Remarkable performance, significantly surpassing the state-of-the-art models by 50%-95% for human preference metrics.
- Detailed illustrations compare various editing scenarios and styles, demonstrating the robustness and adaptability of the system.
- Results of experiments showcase the practical applications and effectiveness of DynVideo-E in maintaining temporal and motion consistency in edited videos.

The poster underscores the importance of advanced video editing tools for enhancing human-centric visual content, signaling a significant leap in 3D video editing technology.
Text transcribed from the image:
NUS | Department of Computer Science
ARC | Applied Research Centre
DynVideo-E: Harnessing Dynamic NeRF for Lang
Jia-Wei Liu^1, Yan-Pei Cao^3, Jay Zhangjie Wu^1, Weijia Ma, 1 Show Lab, National Uni

Highlight: Video-3D Representation for Challenging Human-Centric Video Editing
Problem Definition: Given a human-centric video and a subject and background image, how to edit it?
Challenges: Long-term consistency for large motion and viewpoint changes!

Ref
Subject
Background
Ref Style

Video-3D Representation with a set of image-based 3D dynamic human editing designs.
Significantly outperforms SOTAs on two challenging datasets by 50%~95% for human perference.

Experimental Results

DynVideo-E edits foreground and background with high temporal and motion consistency!