PhysDiff: Physics-Guided Human Motion Diffusion Model

์ €์ž: Ye Yuan, Jiaming Song, Umar Iqbal, Arash Vahdat, Jan Kautz | ๋‚ ์งœ: 2022-12-05 | URL: https://arxiv.org/abs/2212.02500 📄 PDF


Essence

Figure 1

Figure 1. Our PhysDiff model generates physically-plausible motions using a physics-based motion projection in the diffu

PhysDiff๋Š” diffusion ๊ณผ์ •์— ๋ฌผ๋ฆฌ ๊ธฐ๋ฐ˜ motion projection ๋ชจ๋“ˆ์„ ํ†ตํ•ฉํ•˜์—ฌ physically-plausible human motion์„ ์ƒ์„ฑํ•˜๋Š” physics-guided motion diffusion ๋ชจ๋ธ์ด๋‹ค. ๊ธฐ์กด motion diffusion ๋ชจ๋ธ์˜ floating, foot sliding, ground penetration ๊ฐ™์€ ๋ฌผ๋ฆฌ์  artifacts๋ฅผ ์ œ๊ฑฐํ•œ๋‹ค.

Motivation

Achievement

Figure 4

Figure 4. Visual comparison of PhysDiff against the SOTA, MDM [79], on HumanML3D, HumanAct12, and UESTC. PhysDiff reduce

How

Figure 3

Figure 3. Overview of PhysDiff. Each physics-guided diffusion step denoises a motion from timestep t to s, where physics

Originality

Limitation & Further Study

Evaluation

Novelty: 4/5 Technical Soundness: 4/5 Significance: 4/5 Clarity: 4/5 Overall: 4/5

์ดํ‰: PhysDiff๋Š” human motion generation์— physics ์ œ์•ฝ์„ systematically ํ†ตํ•ฉํ•˜์—ฌ physically-plausible motion ์ƒ์„ฑ์˜ ํ•ต์‹ฌ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•œ ํ˜์‹ ์  ์—ฐ๊ตฌ์ด๋‹ค. Iterative projection ์ „๋žต๊ณผ ์ฒ ์ €ํ•œ ์‹คํ—˜ ๋ถ„์„์ด ํ•™๊ณ„์— ์ค‘์š”ํ•œ ๊ธฐ์—ฌ๋ฅผ ์ œ๊ณตํ•˜๋ฉฐ, ์‹ค์ œ animation/VR ์‘์šฉ์˜ ํ˜„์‹คํ™”๋ฅผ ํฌ๊ฒŒ ์•ž๋‹น๊ธด๋‹ค.

← ๋ชฉ๋ก์œผ๋กœ ๋Œ์•„๊ฐ€๊ธฐ

๐ŸŽง Audio Overview

์ด ๋…ผ๋ฌธ ๋ฆฌ๋ทฐ๋ฅผ ํŒŸ์บ์ŠคํŠธํ˜• ์˜ค๋””์˜ค๋กœ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. (Gemini ยท ํ‚ค๋Š” ๋ธŒ๋ผ์šฐ์ €์—๋งŒ ์ €์žฅ ยท ์™„์„ฑ๋ณธ์€ ์ด๋ฉ”์ผ๋กœ๋„ ์ „์†ก)
โ–ธ ๊ณ ๊ธ‰: ๊ตฌ์„ฑ ๋ฐฉํ–ฅ(๋Œ€๋ณธ ์ž‘์„ฑ ์ง€์นจ) ์ง์ ‘ ์ˆ˜์ •