Post

Neural Parametric Gaussian Avatars ๐Ÿ”ด ๐Ÿ”ด ๐Ÿ”ด ( NPGA )

The creation of high-fidelity digital human heads is crucial for integrating virtual elements into daily life, demanding photo-realism and real-time rendering.

Neural Parametric Gaussian Avatars (NPGA) use a data-driven approach with 3D Gaussian splatting for efficient rendering and topological flexibility, conditioning avatar dynamics on neural parametric head models (NPHM).

Evaluations on the NeRSemble dataset show NPGA outperforms previous methods in self-reenactment tasks and demonstrates accurate animations from monocular videos.

๐Ÿง™Paper Authors: Minghan Qin1, Wanhua Li2โ€ , Jiawei Zhou1, Haoqian Wang1โ€ , Hanspeter Pfister2 ( indicates equal contribution, โ€  means Co-corresponding author) 1Tsinghua University, 2Harvard University

Translate to Korean

๊ณ ์ถฉ์‹ค๋„ ๋””์ง€ํ„ธ ํœด๋จผ ํ—ค๋“œ์˜ ์ƒ์„ฑ์€ ๊ฐ€์ƒ ์š”์†Œ๋ฅผ ์ผ์ƒ ์ƒํ™œ์— ํ†ตํ•ฉํ•˜๋Š” ๋ฐ ๋งค์šฐ ์ค‘์š”ํ•˜๋ฉฐ, ํฌํ†  ๋ฆฌ์–ผ๋ฆฌ์ฆ˜๊ณผ ์‹ค์‹œ๊ฐ„ ๋ Œ๋”๋ง์ด ์š”๊ตฌ๋ฉ๋‹ˆ๋‹ค.

NPGA(Neural Parametric Gaussian Avatars)๋Š” ํšจ์œจ์ ์ธ ๋ Œ๋”๋ง๊ณผ ํ† ํด๋กœ์ง€ ์œ ์—ฐ์„ฑ์„ ์œ„ํ•ด 3D ๊ฐ€์šฐ์Šค ์Šคํ”Œ๋ž˜ํŒ…๊ณผ ํ•จ๊ป˜ ๋ฐ์ดํ„ฐ ๊ธฐ๋ฐ˜ ์ ‘๊ทผ ๋ฐฉ์‹์„ ์‚ฌ์šฉํ•˜์—ฌ NPHM(Neural Parametric Head Model)์—์„œ ์•„๋ฐ”ํƒ€ ์—ญํ•™์„ ์กฐ์ ˆํ•ฉ๋‹ˆ๋‹ค.

NeRSemble ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ๋Œ€ํ•œ ํ‰๊ฐ€์— ๋”ฐ๋ฅด๋ฉด NPGA๋Š” ์ž์ฒด ์žฌ์—ฐ ์ž‘์—…์—์„œ ์ด์ „ ๋ฐฉ๋ฒ•์„ ๋Šฅ๊ฐ€ํ•˜๋ฉฐ ๋‹จ์•ˆ ๋น„๋””์˜ค์˜ ์ •ํ™•ํ•œ ์• ๋‹ˆ๋ฉ”์ด์…˜์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.

This post is licensed under CC BY 4.0 by the author.