4DAnyone framework reconstructs 4D humans from single videos

Researchers have developed 4DAnyone, a new framework capable of reconstructing detailed 4D human models from single, uncalibrated monocular videos. This system addresses limitations in existing video diffusion models by employing Reference Context Packing (RCP) and Target Context Routing (TCR) to maintain multiview consistency, which is crucial for 4D Gaussian Splatting (4DGS) reconstruction. The framework was trained on a combination of synthetic and real-world datasets, demonstrating superior performance in both novel-view video generation and downstream 4DGS reconstruction, with strong generalization capabilities. AI

IMPACT Enables more realistic and detailed 4D avatar creation from readily available monocular video.

RANK_REASON This is a research paper detailing a new method for 4D human reconstruction. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• 4DAnyone
• 4D Gaussian Splatting
• 4DGS
• Diffusion Transformer
• DyMVHumans
• MVGameHuman
• Reference Context Packing
• Target Context Routing

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →