This website requires JavaScript.

DINAR: Diffusion Inpainting of Neural Textures for One-Shot Human Avatars

David SvitovDmitrii GudkovRenat BashirovVictor Lemptisky
Mar 2023
摘要
We present DINAR, an approach for creating realistic rigged fullbody avatarsfrom single RGB images. Similarly to previous works, our method uses neuraltextures combined with the SMPL-X body model to achieve photo-realistic qualityof avatars while keeping them easy to animate and fast to infer. To restore thetexture, we use a latent diffusion model and show how such model can be trainedin the neural texture space. The use of the diffusion model allows us torealistically reconstruct large unseen regions such as the back of a persongiven the frontal view. The models in our pipeline are trained using 2D imagesand videos only. In the experiments, our approach achieves state-of-the-artrendering quality and good generalization to new poses and viewpoints. Inparticular, the approach improves state-of-the-art on the SnapshotPeople publicbenchmark.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答