๋ณธ๋ฌธ ๋ฐ”๋กœ๊ฐ€๊ธฐ
Deep Learning/[๋…ผ๋ฌธ] Paper Review

cGAN/Pix2Pix

by ์ œ๋ฃฝ 2023. 7. 7.
728x90
๋ฐ˜์‘ํ˜•

 

 

1. GAN
2. cGAN
  • ์–ด๋–ค ์ˆซ์ž๋ฅผ ๋งŒ๋“ค์–ด๋‚ผ์ง€์— ๋Œ€ํ•œ ์ •๋ณด๋ฅผ ๋„ฃ์–ด์ฃผ๋Š” ๊ฒƒ(์–ด๋–ค ํด๋ž˜์Šค์— ํ•ด๋‹นํ•˜๋Š”์ง€์— ๋Œ€ํ•œ ์ •๋ณด)
  • ex) 7์„ ๋งŒ๋“ค๊ณ ์ž ํ•œ๋‹ค๋ฉด condition vector์— 7์„ ๋„ฃ์–ด์ฃผ๊ณ , z(noise)์—๋Š” ๋žœ๋คํ•˜๊ฒŒ ์ƒ˜ํ”Œ๋งํ•ด์„œ 7์ด๋ผ๋Š” ์˜๋ฏธ๋ฅผ ๊ฐ€์ง€๋Š” ๋žœ๋คํ•œ ํ˜•ํƒœ๋ฅผ ๋งŒ๋“ค์–ด์คŒ
3. Pix2Pix
  • image to image translation : ์ด๋ฏธ์ง€์˜ ํŠน์ • ์–‘์ƒ์„ ๋‹ค๋ฅธ ์–‘์ƒ์œผ๋กœ ๋ฐ”๊ฟ”์ฃผ๋Š” ๊ฒƒ์„ ์˜๋ฏธ
  • ex) ์†๊ทธ๋ฆผ → ์‹ค์ œ ์‚ฌ์ง„์œผ๋กœ translation
  • ์ด๋ฏธ์ง€ ์ž์ฒด๋ฅผ condition์œผ๋กœ ๋ฐ›์•„๋ฒ„๋ฆผ (์ด๋ฏธ์ง€ ์ž์ฒด๊ฐ€ ์ •๋ณด๊ฐ€ ๋˜๋Š” ๊ฒƒ์ž„)
  • ์ฆ‰, ์ด๋ฏธ์ง€๊ฐ€ ๋“ค์–ด์™”์„ ๋•Œ, ๊ฑฐ๊ธฐ์— ๋ถ€ํ•ฉํ•˜๋Š” output์˜ ํ˜•ํƒœ๋กœ ๋งŒ๋“ค์–ด์คŒ
  • noise vector z๋ฅผ ์•ˆ์”€
  • pixel ์„ ์ž…๋ ฅ์œผ๋กœ ๋ฐ›์•„ pixel์„ ์˜ˆ์ธกํ•จ (์ด๋ฏธ์ง€๋ฅผ ์ด๋ฏธ์ง€๋กœ ๋ฐ˜ํ™˜)
  • ์–˜๋„ค๋“ค์€ paired dataset (์• ์ดˆ์— ์ •๋‹ต ์ด๋ฏธ์ง€๊ฐ€ ๋ญ”์ง€๋ฅผ ์•Œ๊ณ  ํ•™์Šต์„ ์‹œํ‚ด)
  • ์ •๋‹ต์„ ์•Œ๊ณ  ์žˆ๊ธฐ ๋•Œ๋ฌธ์—, ์‹ค์ œ ์ •๋‹ต๊ณผ ์œ ์‚ฌํ•˜๋„๋ก ํ•˜๋Š” loss(L1)์„ ์‚ฌ์šฉ

⇒ ๋‹จ์ : ์„œ๋กœ ๋‹ค๋ฅธ ๋‘ ๋„๋ฉ”์ธ x,y์˜ ๋ฐ์ดํ„ฐ ๋‘๊ฐœ๋ฅผ ํ•œ์Œ์œผ๋กœ ๋ฌถ์–ด์„œ ํ•™์Šต์„ ์ง„ํ–‰์‹œํ‚ด (์†๊ทธ๋ฆผ๊ณผ ๊ทธ์— ๋งž๋Š” ์‚ฌ์ง„-condition)

⇒ ์˜ˆ๋ฅผ ๋“ค์–ด ์†๊ทธ๋ฆผ ์‹ ๋ฐœ๊ณผ ์‹ ๋ฐœ ์‚ฌ์ง„, ์ด๋ ‡๊ฒŒ ๋งค์นญ๋˜์ง€ ์•Š๊ณ , x๋Š” ๊ฑด๋ฌผ์‚ฌ์ง„, y๋Š” ํ’๊ฒฝ์‚ฌ์ง„์œผ๋กœ ๋ฌถ์—ฌ์žˆ๋Š” ๋ฐ์ดํ„ฐ ์…‹์— ๋Œ€ํ•ด์„œ๋„ ๊ณผ์—ฐ ์ ์šฉ์ด ๊ฐ€๋Šฅํ• ๊นŒ?์— ๋Œ€ํ•œ ์˜๋ฌธ์—์„œ ๋‚˜์˜จ ๊ฒƒ์ด cycleGAN์ž„.

728x90
๋ฐ˜์‘ํ˜•

'Deep Learning > [๋…ผ๋ฌธ] Paper Review' ์นดํ…Œ๊ณ ๋ฆฌ์˜ ๋‹ค๋ฅธ ๊ธ€

YOLOv4: Optimal Speed and Accuracy of Object Detection  (0) 2023.07.09
EfficientNet  (0) 2023.07.07
R-CNN  (0) 2023.07.06
GAN: Generative Adversarial Nets  (0) 2023.07.06
AE  (0) 2023.07.06