์ „์ฒด ๊ธ€ 36

2022 KAKAO BLIND RECRUITEMENT Lv1. ์‹ ๊ณ ๋ฐ›๊ธฐ

2022 KAKAO BLIND RECRUITEMENT Lv1. ์‹ ๊ณ ๋ฐ›๊ธฐ ๋ฌธ์ œ : https://programmers.co.kr/learn/courses/30/lessons/92334 ํ”„๋กœ๊ทธ๋ž˜๋จธ์Šค ์ฝ”๋“œ ์ค‘์‹ฌ์˜ ๊ฐœ๋ฐœ์ž ์ฑ„์šฉ. ์Šคํƒ ๊ธฐ๋ฐ˜์˜ ํฌ์ง€์…˜ ๋งค์นญ. ํ”„๋กœ๊ทธ๋ž˜๋จธ์Šค์˜ ๊ฐœ๋ฐœ์ž ๋งž์ถคํ˜• ํ”„๋กœํ•„์„ ๋“ฑ๋กํ•˜๊ณ , ๋‚˜์™€ ๊ธฐ์ˆ  ๊ถํ•ฉ์ด ์ž˜ ๋งž๋Š” ๊ธฐ์—…๋“ค์„ ๋งค์นญ ๋ฐ›์œผ์„ธ์š”. programmers.co.kr ๐Ÿ”ธ ๋‚ด๊ฐ€ ์ƒ๊ฐํ•œ ์ ‘๊ทผ ๋ฐฉ๋ฒ• ์œ ์ €๋ณ„ ์‹ ๊ณ ๋‹นํ•œ ํšŸ์ˆ˜๋ฅผ ์„ธ๋Š” ๋ฆฌ์ŠคํŠธ๋ฅผ ํ•˜๋‚˜ ์ž‘์„ฑ. ํ•ด๋‹น ๋ฆฌ์ŠคํŠธ๋กœ ์ •์ง€๋œ ์‚ฌ์šฉ์ž๋ฅผ ๊ฑฐ๋ฆ„. ์ •์ง€๋œ ์‚ฌ์šฉ์ž๋ฅผ ์‹ ๊ณ ํ•˜๋ฉด answer+1 def solution(id_list, report, k): answer = [0] * len(id_list) block = {x : 0 for x in id_list} for..

Cycle-Consistent Adversarial Networks ๋…ผ๋ฌธ๋ฆฌ๋ทฐ #3

Formulation ๋ชฉํ‘œ๋Š” training sample xโˆˆX, yโˆˆY ์ด ์ฃผ์–ด์ง„ domain X์™€ Y๋ฅผ mapping functions์„ ํ•™์Šตํ•˜๋Š” ๊ฒƒ์œผ๋กœ ๋ชจ๋ธ์—๋Š” ๋‘๊ฐ€์ง€์˜ mapping G, F๊ฐ€ ํฌํ•จ๋˜์–ด์žˆ์Œ(G:Xโ†’Y, F:Yโ†’X) ๊ทธ๋ฆฌ๊ณ  ๋‘ ๊ฐ€์ง€ discriminator Dx์™€ Dy๋ฅผ ์‚ฌ์šฉํ•จ Dx๋Š” ์ด๋ฏธ์ง€ x์™€ ๋ณ€ํ™˜๋œ ์ด๋ฏธ์ง€ F(y)๋ฅผ ๊ตฌ๋ณ„, Dy๋Š” y์™€ G(x)๋ฅผ ๊ตฌ๋ณ„ โ‡’ ์ƒ์„ฑ๋œ ์ด๋ฏธ์ง€์˜ ๋ถ„ํฌ์™€ target domain์˜ ๋ฐ์ดํ„ฐ ๋ถ„ํฌ๋ฅผ ์ผ์น˜์‹œํ‚ค ์œ„ํ•ด adversarial losses ์‚ฌ์šฉ โ‡’ ํ•™์Šต๋œ ๋งคํ•‘ G์™€ F๊ฐ€ ์„œ๋กœ ๋ชจ์ˆœ๋˜๋Š” ๊ฒƒ์„ ๋ง‰๊ธฐ ์œ„ํ•ด consistency losses ์‚ฌ์šฉ Adversarial Loss G๋Š” domain Y์—์„œ ์ด๋ฏธ์ง€์™€ ์œ ์‚ฌํ•œ G(x)๋ฅผ ์ƒ์„ฑํ•˜๋ ค๊ณ  ํ•จ Dy๋Š” ๋ณ€ํ™˜๋œ G..

4. ๋ฐ˜๋ณต๋ฌธ

(1) ๋ฆฌ์ŠคํŠธ์™€ ๋ฐ˜๋ณต๋ฌธ list (๋ฆฌ์ŠคํŠธ) : ์—ฌ๋Ÿฌ๊ฐ€์ง€ ์ž๋ฃŒ๋ฅผ ์ €์žฅํ•  ์ˆ˜ ์žˆ๋Š” ์ž๋ฃŒ element (์š”์†Œ) : ๋Œ€๊ด„ํ˜ธ ๋‚ด๋ถ€์— ๋„ฃ๋Š” ์ž๋ฃŒ index (์ธ๋ฑ์Šค) : ๋ฆฌ์ŠคํŠธ์˜ ์ˆœ์„œ [์š”์†Œ, ์š”์†Œ, ์š”์†Œ ... ] ๋ฆฌ์ŠคํŠธ๋Š” ๋Œ€๊ด„ํ˜ธ[ ]์— ์ž๋ฃŒ๋ฅผ ์‰ผํ‘œ๋กœ ๊ตฌ๋ถ„ํ•ด์„œ ์ž…๋ ฅํ•จ ๋ฆฌ์ŠคํŠธ์˜ ์ธ๋ฑ์Šค ๋˜ํ•œ ๋ฌธ์ž์—ด๊ณผ ๊ฐ™์ด 0๋ถ€ํ„ฐ ์ˆœ์„œ๋Œ€๋กœ ์…ˆ ๋ฆฌ์ŠคํŠธ์˜ ๋‹ค์–‘ํ•œ ์‚ฌ์šฉ๋ฒ• ๋Œ€๊ด„ํ˜ธ ์•ˆ์— ์Œ์ˆ˜๋ฅผ ๋„ฃ์–ด ๋’ค์—์„œ๋ถ€ํ„ฐ ์š”์†Œ๋ฅผ ์„ ํƒํ•  ์ˆ˜ ์žˆ์Œ ์ด์˜ ๊ฒฝ์šฐ ๋งจ ๋’ค์˜ ์š”์†Œ์˜ ์ธ๋ฑ์Šค ๊ฐ’์ด -1, ์ œ์ผ ์ฒซ๋ฒˆ์žฌ ์š”์†Œ์˜ ์ธ๋ฑ์Šค ๊ฐ’์ด -n์ด ๋จ ๋ฆฌ์ŠคํŠธ ์ ‘๊ทผ ์—ฐ์‚ฐ์ž๋ฅผ ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ด์ค‘์œผ๋กœ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Œ list[3] : ๋ฆฌ์ŠคํŠธ์˜ 3๋ฒˆ์งธ ์š”์†Œ list[3][0] : ๋ฆฌ์ŠคํŠธ์˜ 3๋ฒˆ์งธ ์š”์†Œ์˜ 0๋ฒˆ์งธ ์š”์†Œ (๋ฌธ์ž์—ด์˜ ๊ฒฝ์šฐ ์ฒซ๋ฒˆ์งธ ๋ฌธ์ž) ๋ฆฌ์ŠคํŠธ ์•ˆ์— ๋ฆฌ์ŠคํŠธ๋ฅผ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Œ..

Cycle-Consistent Adversarial Networks ๋…ผ๋ฌธ๋ฆฌ๋ทฐ #2

Related work ๋ณ€ํ™˜ํ•œ ์ด๋ฏธ์ง€์™€ target domain ์˜ ์ด๋ฏธ์ง€์™€ ๊ตฌ๋ณ„ํ•  ์ˆ˜ ์—†๋„๋ก ๋งคํ•‘์„ ํ•™์Šตํ•˜๊ธฐ์œ„ํ•ด adversarial loss๋ฅผ ์‚ฌ์šฉํ•จ Image-to-Image Translation ์ด๋“ค์˜ ์ด๋ฏธ์ง€ ๋ณ€ํ™˜์€ Isola๋“ฑ์˜ pix2pix์˜ framework๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•จ. conditional generative adversarial network ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ input-output ์ด๋ฏธ์ง€์˜ mapping ์„ ํ•™์Šตํ•จ Unpaired Image-to-Image Rosales๋“ฑ ๋‘ ๋ฐ์ดํ„ฐ์˜ domain์„ ์—ฐ๊ด€์‹œํ‚ค๋Š” ๊ฒƒ์ด ๋ชฉํ‘œ. ์†Œ์Šค์ด๋ฏธ์ง€์—์„œ ๊ณ„์‚ฐํ•œ Markov random field ์™€ ๋‹ค์–‘ํ•œ ์Šคํƒ€์ผ์˜ ์ด๋ฏธ์ง€์—์„œ ์–ป์€ likelihood term ์„ ํฌํ•จํ•˜๋Š” Bayesian framework๋ฅผ..

Cycle-Consistent Adversarial Networks ๋…ผ๋ฌธ๋ฆฌ๋ทฐ #1

Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks Abstract Image-to-Image ๋ณ€ํ™˜์€ ์ด๋ฏธ์ง€ ์Œ์˜ ์„ธํŠธ๋ฅผ ์‚ฌ์šฉํ•ด์„œ ์ž…๋ ฅ ์ด๋ฏธ์ง€์™€ ์ถœ๋ ฅ ์ด๋ฏธ์ง€ ์‚ฌ์ด์˜ ๋งคํ•‘์„ ํ•™์Šตํ•˜๋Š” ๊ฒƒ์ด ๋ชฉํ‘œ โ†’ ๋Œ€๋ถ€๋ถ„์˜ ์ž‘์—…์˜ ๊ฒฝ์šฐ, ์Œ์œผ๋กœ ๋ชจ๋“  ๋ฐ์ดํ„ฐ๋ฅผ ํ•™์Šต ํ•  ์ˆ˜ ์—†๋‹ค๋Š” ํ•œ๊ณ„์  ์กด์žฌ ์Œ์œผ๋กœ ๋œ ์˜ˆ์‹œ๊ฐ€ ์—†๋Š” ๊ฒฝ์šฐ, X โ†’ Y ๋กœ ์ด๋ฏธ์ง€๋ฅผ ๋ณ€ํ™˜ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ํ•™์Šตํ•˜๊ธฐ ์œ„ํ•œ ๋ฐฉ๋ฒ• ์ œ์‹œ โ†’ G:Xโ†’Y ๋ฅผ ์—ญ๋งคํ•‘ F:Yโ†’X ์™€ ๊ฒฐํ•ฉ. F(G(X))โ‰ˆX ์ด ์ ์šฉ๋˜๊ธฐ ์œ„ํ•ด a cycle consistency loss ๋„์ž… Introduction ์™ผ์ชฝ์˜ ์‚ฌ์ง„๋“ค์€ Paired data ์ด๋‚˜, ์˜ค๋ฅธ์ชฝ์˜ ์‚ฌ์ง„๋“ค์€ Unpaired data ๊ตฌ์„ฑ์ฒด ์„œ๋กœ..

3. ์กฐ๊ฑด๋ฌธ

(1) ๋ถˆ ์ž๋ฃŒํ˜•๊ณผ if ์กฐ๊ฑด๋ฌธ Boolean ๋ถˆ๋ฆฐ, ๋ถˆ๋ฆฌ์–ธ ํ”„๋กœ๊ทธ๋ž˜๋ฐ์—์„œ๋Š” ์งง๊ฒŒ Bool ๋ถˆ์ด๋ผ๊ณ  ํ‘œํ˜„ํ•จ : ์˜ค์ง True(์ฐธ), False(๊ฑฐ์ง“) ๊ฐ’๋งŒ ๊ฐ€์งˆ ์ˆ˜ ์žˆ์Œ ๋ถˆ ๋งŒ๋“ค๊ธฐ : ๋น„๊ต ์—ฐ์‚ฐ์ž == : ๊ฐ™๋‹ค != : ๋‹ค๋ฅด๋‹ค : ํฌ๋‹ค = : ํฌ๊ฑฐ๋‚˜ ๊ฐ™๋‹ค < : ์ž‘๋‹ค