Sign In

Physics-Based Benchmarking Metrics for Multimodal Synthetic Images

์ž‘์„ฑ์ž
  • Haebom
์นดํ…Œ๊ณ ๋ฆฌ
Empty

์ €์ž

Kishor Datta Gupta, Marufa Kamal, Md. Mahfuzur Rahman, Fahad Rahman, Mohd Ariful Haque, Sunzida Siddique

๐Ÿ’ก ๊ฐœ์š”

๊ธฐ์กด์˜ ๋‹ค์ค‘๋ชจ๋‹ฌ ํ•ฉ์„ฑ ์ด๋ฏธ์ง€ ํ‰๊ฐ€ ์ง€ํ‘œ๋“ค์€ ์ „๋ฌธ ๋ถ„์•ผ๋‚˜ ํŠน์ • ๋งฅ๋ฝ์—์„œ ์˜๋ฏธ๋ก ์ , ๊ตฌ์กฐ์  ์ •ํ™•์„ฑ์„ ํฌ์ฐฉํ•˜๋Š” ๋ฐ ํ•œ๊ณ„๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด์— ๋ณธ ๋…ผ๋ฌธ์€ ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ, ์ง€์‹ ๊ธฐ๋ฐ˜ ๋งคํ•‘, ์‹œ๊ฐ-์–ธ์–ด ๋ชจ๋ธ์„ ๊ฒฐํ•ฉํ•œ ๋ฌผ๋ฆฌ ๊ธฐ๋ฐ˜ ํ‰๊ฐ€ ์ง€ํ‘œ(PCMDE)๋ฅผ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ์ด ์ง€ํ‘œ๋Š” ๊ฐ์ฒด ํƒ์ง€ ๋ฐ VLM์„ ํ†ตํ•œ ๊ณต๊ฐ„-์„ฑ๋Šฅ ํŠน์ง• ์ถ”์ถœ, ์ ์‘ํ˜• ๊ตฌ์„ฑ ์š”์†Œ ์ˆ˜์ค€ ๊ฒ€์ฆ์„ ์œ„ํ•œ ์‹ ๋ขฐ๋„ ๊ฐ€์ค‘์น˜ ๊ธฐ๋ฐ˜ ์œตํ•ฉ, ๊ทธ๋ฆฌ๊ณ  LLM์„ ํ™œ์šฉํ•œ ๋ฌผ๋ฆฌ ๊ธฐ๋ฐ˜ ์ถ”๋ก ์„ ํ†ตํ•ด ์ด๋Ÿฌํ•œ ์ œ์•ฝ์„ ๊ทน๋ณตํ•ฉ๋‹ˆ๋‹ค.

๐Ÿ”‘ ์‹œ์‚ฌ์  ๋ฐ ํ•œ๊ณ„

โ€ข
๋‹ค์ค‘๋ชจ๋‹ฌ ํ•ฉ์„ฑ ์ด๋ฏธ์ง€์˜ ์˜๋ฏธ๋ก ์  ๋ฐ ๊ตฌ์กฐ์  ์ •ํ™•์„ฑ์„ ๋” ์ž˜ ํฌ์ฐฉํ•˜๋Š” ์ƒˆ๋กœ์šด ๋ฌผ๋ฆฌ ๊ธฐ๋ฐ˜ ํ‰๊ฐ€ ์ง€ํ‘œ๋ฅผ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค.
โ€ข
๊ธฐ์กด ์ง€ํ‘œ์˜ ํ•œ๊ณ„๋ฅผ ๊ทน๋ณตํ•˜๊ณ , ํŠนํžˆ ์ „๋ฌธ ๋ถ„์•ผ๋‚˜ ํŠน์ • ๋งฅ๋ฝ์—์„œ ๋ณด๋‹ค ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๋Š” ํ‰๊ฐ€๋ฅผ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค.
โ€ข
์ œ์•ˆ๋œ PCMDE ์ง€ํ‘œ์˜ ์‹ค์ œ ์ ์šฉ ๊ฐ€๋Šฅ์„ฑ ๋ฐ ๋‹ค๋ฅธ ๋‹ค์ค‘๋ชจ๋‹ฌ ์ƒ์„ฑ ์ž‘์—…์œผ๋กœ์˜ ํ™•์žฅ์„ฑ์— ๋Œ€ํ•œ ์ถ”๊ฐ€์ ์ธ ์—ฐ๊ตฌ๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
๐Ÿ‘