์ ์: Ye Yuan, Jiaming Song, Umar Iqbal, Arash Vahdat, Jan Kautz | ๋ ์ง: 2022-12-05 | URL: https://arxiv.org/abs/2212.02500 📄 PDF
Figure 1. Our PhysDiff model generates physically-plausible motions using a physics-based motion projection in the diffu
PhysDiff๋ diffusion ๊ณผ์ ์ ๋ฌผ๋ฆฌ ๊ธฐ๋ฐ motion projection ๋ชจ๋์ ํตํฉํ์ฌ physically-plausible human motion์ ์์ฑํ๋ physics-guided motion diffusion ๋ชจ๋ธ์ด๋ค. ๊ธฐ์กด motion diffusion ๋ชจ๋ธ์ floating, foot sliding, ground penetration ๊ฐ์ ๋ฌผ๋ฆฌ์ artifacts๋ฅผ ์ ๊ฑฐํ๋ค.
Figure 4. Visual comparison of PhysDiff against the SOTA, MDM [79], on HumanML3D, HumanAct12, and UESTC. PhysDiff reduce
Figure 3. Overview of PhysDiff. Each physics-guided diffusion step denoises a motion from timestep t to s, where physics
์ดํ: PhysDiff๋ human motion generation์ physics ์ ์ฝ์ systematically ํตํฉํ์ฌ physically-plausible motion ์์ฑ์ ํต์ฌ ๋ฌธ์ ๋ฅผ ํด๊ฒฐํ ํ์ ์ ์ฐ๊ตฌ์ด๋ค. Iterative projection ์ ๋ต๊ณผ ์ฒ ์ ํ ์คํ ๋ถ์์ด ํ๊ณ์ ์ค์ํ ๊ธฐ์ฌ๋ฅผ ์ ๊ณตํ๋ฉฐ, ์ค์ animation/VR ์์ฉ์ ํ์คํ๋ฅผ ํฌ๊ฒ ์๋น๊ธด๋ค.