๐Ÿค– Computer Vision์ด๋ž€? | ๋‚ด๊ฐ€๋ณด๋ ค๊ณ ์ •๋ฆฌํ•œAI๐Ÿง

HipJaengYiCatยท2023๋…„ 3์›” 31์ผ
0

DeepLearning

๋ชฉ๋ก ๋ณด๊ธฐ
9/16
post-thumbnail

preview

์‚ฌ๋žŒ์€ ์‹œ๊ฐ, ์ฒญ๊ฐ ๋“ฑ๊ณผ ๊ฐ™์ด ์˜ค๊ฐ์„ ํ†ตํ•ด ์„ธ์ƒ๊ณผ ์ƒํ˜ธ์ž‘์šฉ์„ ํ•˜๋ฉด์„œ ์„ฑ์žฅํ•œ๋‹ค. ์‚ฌ๋žŒ์ด ๊ฐ๊ฐ์„ ํ†ตํ•ด ๋ฐ›์•„๋“ค์ด๋Š” ์ •๋ณด์˜ 75%๋Š” ์‹œ๊ฐ์„ ํ†ตํ•ด ์˜จ๋‹ค๊ณ  ํ•œ๋‹ค.
๋”ฐ๋ผ์„œ ์ธ๊ฐ„์˜ ์ง€๋Šฅ์„ ๋ชจ๋ฐฉํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ์‹œ๊ฐ์„ ๋ชจ๋ฐฉํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•  ๊ฒƒ์ด๋‹ค.
์ด๋Ÿฐ ์ธ๊ฐ„์˜ ์‹œ๊ฐ์„ ๋ชจ๋ฐฉํ•˜๋Š” ๊ฒƒ์ด Computer Vision์ด๋ผ๊ณ  ํ•  ์ˆ˜ ์žˆ๋‹ค.

Computer Vision์ด๋ž€?

์ปดํ“จํ„ฐ ๋น„์ „์ด๋ž€?
์ปดํ“จํ„ฐ ๋น„์ „์€ ์‹œ๊ฐ์  ์„ธ๊ณ„๋ฅผ ํ•ด์„ํ•˜๊ณ  ์ดํ•ดํ•˜๋„๋ก ์ปดํ“จํ„ฐ๋ฅผ ํ•™์Šต์‹œํ‚ค๋Š” ์ธ๊ณต ์ง€๋Šฅ ๋ถ„์•ผ์ž…๋‹ˆ๋‹ค. ์ปดํ“จํ„ฐ๊ฐ€ ์นด๋ฉ”๋ผ์™€ ๋™์˜์ƒ์—์„œ ๋””์ง€ํ„ธ ์ด๋ฏธ์ง€์™€ ๋”ฅ ๋Ÿฌ๋‹ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐ์ฒด๋ฅผ ์ •ํ™•ํ•˜๊ฒŒ ์‹๋ณ„ํ•˜๊ณ  ๋ถ„๋ฅ˜ํ•˜๋Š” ํ•™์Šต์„ ๋งˆ์น˜๋ฉด "๊ด€์ฐฐ" ๋Œ€์ƒ์— ๋ฐ˜์‘ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

์ฆ‰, ์ปดํ“จํ„ฐ์—์„œ ์–ด๋–ป๊ฒŒ ๋ณด๊ณ (visual perception) ์ƒ์ƒํ•˜๋Š”์ง€(visual intelligence)๋ฅผ ๊ฐ€๋ฅด์น˜๋Š” ๊ฒƒ์œผ๋กœ ๋ณผ ์ˆ˜ ์žˆ๋‹ค.

๐Ÿ’โ€โ™€๏ธ ์ธ๊ฐ„์ด ์‹œ๊ฐ์ ์œผ๋กœ ๋ฐ›์•„๋“ค์ด๋Š” ๋ฐ์ดํ„ฐ๋Š” ๋ฌด์—‡์ผ๊นŒ?
์ธ๊ฐ„์˜ ์‹œ๊ฐ์  ์„ธ๊ณ„๋ฅผ ๋ฐ์ดํ„ฐ๋กœ ํ‘œํ˜„ํ•˜๋ฉด visual data๋ผ๊ณ  ํ• ์ˆ˜ ์žˆ๋‹ค.
visual data : image data ๋˜๋Š” video data ์ด๋‹ค

๐Ÿ’โ€โ™€๏ธ ๊ทธ๋ ‡๋‹ค๋ฉด ์ธ๊ฐ„์˜ ์‹œ๊ฐ์  ์ธ์ง€๋Š” ์–ด๋–ค ๊ฒƒ๋“ค์ด ์žˆ์„๊นŒ?

  • color perception

  • motion perception

  • 3D perception

  • semantic-level perception

  • social perception(emotion perception)

  • visuomotor perception
    - ๋™์ผํ•œ ๋‘ ๊ฐœ์ฒด๊ฐ€ ๊ฐ™์€์ง€ ์‹๋ณ„ํ•˜๋Š” Visual discrimination
    - ์‹œ๊ฐ ์ •๋ณด๋ฅผ ๊ธฐ์–ตํ•˜๋Š” Visual memory
    - ๋‘ ๊ฐœ์ฒด๊ฐ€ ํฌ๊ธฐ, ์ƒ‰์ƒ์ด ๋‹ฌ๋ผ๋„ ๊ฐ™๋‹ค๋Š” ๊ฒƒ์„ ์•„๋Š” Form constancy
    - ๋ณต์žกํ•œ ๋ฐฐ๊ฒฝ์— ๋ฌผ์ฒด๊ฐ€ ์ˆจ๊ฒจ์ ธ ์žˆ์„๋•Œ ๋ฌผ์ฒด๋ฅผ ์ฐพ๋Š” Figure ground
    - ๊ฐœ์ฒด ์ค‘ ํ•˜๋‚˜์˜ ์ผ๋ถ€๊ฐ€ ๋ˆ„๋ฝ๋˜๋”๋ผ๋„ ๋™์ผํ•œ ๋‘ ๊ฐœ์ฒด๋ฅผ ์‹๋ณ„ํ•˜๋Š” Visual closure
    - ์œ„์˜ ๊ฐœ๋…๋“ค์„ ํฌํ•จํ•˜๋Š” ์ธ์ง€๋Šฅ๋ ฅ์œผ๋กœ ์†๊ธ€์”จ๋ฅผ ์ฝ๊ฑฐ๋‚˜, ๋ฏธ๋กœ๋ฅผ ํƒˆ์ถœํ•˜๊ฑฐ๋‚˜, ํผ์ฆ ๋งž์ถ”๊ธฐ ๋“ฑ์„ ํ•˜๋Š” ๋Šฅ๋ ฅ์œผ๋กœ๋„ ๋ณผ์ˆ˜ ์žˆ๋‹ค.

  • etc(understanding human visual perception capability)

CV์˜ tasks

๐Ÿ’โ€โ™€๏ธ ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ์ˆ ์ด ๋‚˜์˜ค๊ธฐ ์ „์— CV๋Š” ์–ด๋–ป๊ฒŒ ๊ตฌํ˜„ํ–ˆ์„๊นŒ?

* ๊ธฐ์กด์—๋Š” ์‹œ๊ฐ ๋ฐ์ดํ„ฐ๋ฅผ ์ธ๊ฐ„์ด ์ง์ ‘ ํŠน์ง•์„ ์ถ”์ถœํ•ด ํ•™์Šต์„ ์‹œ์ผฐ๋‹ค. ์ธ๊ฐ„์ด ์ง์ ‘ ํŠน์ง•์„ ์ถ”์ถœํ•˜๋‹ค๋ณด๋‹ˆ ์ธ์ ์ž์› ์†Œ๋ชจ๋„ ๋†’๊ณ  ์ธ๊ฐ„์˜ ํŽธํ–ฅ์ด ๋“ค์–ด๊ฐ€ ๋ถˆ์•ˆ์ •ํ•œ ๊ฒฐ๊ณผ๋ฌผ์„ ๋„์ถœํ•ด๋ƒˆ๋‹ค

๐Ÿ’โ€โ™€๏ธ ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ์ˆ ์ด ์–ด๋–ป๊ฒŒ CV๋Š” ๋ฐ”๊พธ์—ˆ์„๊นŒ?

* ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ์ˆ ์ด ๋‚˜์˜จ ์ดํ›„ ์ธ๊ฐ„์ด ์ง์ ‘ ํŠน์ง•์„ ์ถ”์ถœํ•˜์ง€ ์•Š๊ณ  ์ปดํ“จํ„ฐ๊ฐ€ ํŠน์ง• ์ถ”์ถœ๊ณผ ๋ถ„๋ฅ˜๋ฅผ ์ฒ˜๋ฆฌํ•ด ๋” ๋งŽ์€ ๋ฐ์ดํ„ฐ๋ฅผ ์ฒ˜๋ฆฌํ•  ์ˆ˜ ์žˆ๊ฒŒ ๋˜์—ˆ๋‹ค.

๐Ÿ’โ€โ™€๏ธ ๊ทธ๋ ‡๋‹ค๋ฉด CV๋Š” ์–ด๋–ค ๊ณผ์ œ๋“ค์ด ์žˆ์„๊นŒ?

  • classification

  • semantic segmentaion

  • object detection

  • instance segmentation & panoptic segmentation

  • data augmentation

  • knowledge distillation
    ํฐ ๋ชจ๋ธ์˜ ์ง€์‹์„ ์ž‘์€ ๋ชจ๋ธ๋กœ transferํ•จ

  • multi model
    vision data ๋ฟ๋งŒ์•„๋‹ˆ๋ผ text, sound data๋„ ๋‹ค๋ฃฐ ์ˆ˜ ์žˆ๋Š” ๋ชจ๋ธ

  • conditional generative model(์ด๋ฏธ์ง€ ์ƒ์„ฑ๋ชจ๋ธ)

  • neural network analysis by visualization
    neural network์˜ ์‹œ๊ฐ์  ์ดํ•ด์™€ ๋””๋ฒ„๊น… ์‹œ๊ฐํ™”

Epilogue

๐Ÿ’โ€โ™€๏ธ ๋‹ค์Œ ์žฅ์—์„œ๋Š” CV์˜ ์‹œ์ ์ด๋ผ๊ณ  ๋ณผ ์ˆ˜ ์žˆ๋Š” classification ๋ถ€ํ„ฐ ์‹œ์ž‘ํ•ด์„œ ์•ž์—์„œ ๋‹ค๋ฃฌ CV์˜ ๊ณผ์ œ๋“ค์„ ํ•˜๋‚˜ ํ•˜๋‚˜์”ฉ ๋‹ค๋ฃฐ ์˜ˆ์ •์ด๋‹ค.

-์ฐธ๊ณ  -

๋„ค์ด๋ฒ„ ๋ถ€์ŠคํŠธ์บ ํ”„ AITech 5๊ธฐ ์ž๋ฃŒ
https://www.sas.com/ko_kr/insights/analytics/computer-vision.html

profile
AI Learning, Parcelled Innovations, Carrying All

0๊ฐœ์˜ ๋Œ“๊ธ€