DreamWaltz: Make a Scene with Complex 3D Animatable AvatarsWe present DreamWaltz, a novel framework for generating and animating complex 3D avatars given text guidance and parametric human body prior. While recent methods have shown encouraging results for text-to-3D generation of common objects, creating high-quaidea-research.github.io DreamWaltz: Make a Scene with Complex 3D Animatable Avatars..

HumanSplat: Generalizable Single-Image Human Gaussian Splatting with Structure PriorsDespite recent advancements in high-fidelity human reconstruction techniques, the requirements for densely captured images or time-consuming per-instance optimization significantly hinder their applications in broader scenarios. To tackle these issues, wearxiv.org HumanSplat: Generalizable Single-Image Human Gau..

GitHub - yukangcao/JIFFContribute to yukangcao/JIFF development by creating an account on GitHub.github.com  JIFF: Jointly-aligned Implicit Face Function for High Quality Single View Clothed Human ReconstructionThis paper addresses the problem of single view 3D human reconstruction. Recent implicit function based methods have shown impressive results, but they fail to recover fine face details i..

DoubleField Project PageWe introduce DoubleField, a novel framework combining the merits of both surface field and radiance field for high-fidelity human reconstruction and rendering. Within DoubleField, the surface field and radiance field are associated together by a shared feawww.liuyebin.com  DoubleField: Bridging the Neural Surface and Radiance Fields for High-fidelity Human Reconstruction ..

Abstract복잡한 pose를 취하고 있는 사람이나 옷이나 헤어스타일 등을 리얼하게 복구하는 것은 보이지 않는 영역을 예측하는 것 뿐만 아니라 중요한 task들 중 하나로 여겨져 왔다. 하지만 이전의 모델들은 2D image를 3D로 변환하고 texture를 예측하는 것에 있어서 prior guidance가 충분하지 않다는 점이 문제가 되어왔다. 따라서 본 논문에서는 SIFU [Side-view Conditioned Implicit Function for Real-world Usable Clothed Human Reconstruction]이라는 모델을 제안해서 이를 해결하고자 했다.SIFU는 transformer의 cross-mechanism을 사용하였고, SMPL-X를 이용해서 2D feature들을..

Abstractimage-based 3D human shape estimation은 Deep Neural Network의 등장으로 급격히 발전했다. 하지만, real world setting에서는 input image의 detail을 살리는데 어려움을 겪는데, 이 논문의 저자들은 이러한 어려움의 원인을 2가지 conflicting requirements에서 찾았다.Accurate predictions require large context, but precise predictions require high resolutionDue to memory limitations in current hardware, previous approaches tend to take low resolution images ..