์ฝ˜ํ…์ธ ๋กœ ๊ฑด๋„ˆ๋›ฐ๊ธฐ

๋‹ค์Œ์— ๋Œ€ํ•œ ํฌ๊ด„์ ์ธ ์ž์Šต์„œ Ultralytics YOLO

Ultralytics' YOLO ๐Ÿš€ ๊ฐ€์ด๋“œ์— ์˜ค์‹  ๊ฒƒ์„ ํ™˜์˜ํ•ฉ๋‹ˆ๋‹ค! ํฌ๊ด„์ ์ธ ํŠœํ† ๋ฆฌ์–ผ์€ ํ›ˆ๋ จ ๋ฐ ์˜ˆ์ธก์—์„œ ๋ฐฐํฌ์— ์ด๋ฅด๊ธฐ๊นŒ์ง€ YOLO ๊ฐ์ฒด ๊ฐ์ง€ ๋ชจ๋ธ์˜ ๋‹ค์–‘ํ•œ ์ธก๋ฉด์„ ๋‹ค๋ฃน๋‹ˆ๋‹ค. PyTorch ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ตฌ์ถ•๋œ YOLO ์€ ์‹ค์‹œ๊ฐ„ ๊ฐ์ฒด ๊ฐ์ง€ ์ž‘์—…์—์„œ ํƒ์›”ํ•œ ์†๋„์™€ ์ •ํ™•์„ฑ์„ ์ž๋ž‘ํ•ฉ๋‹ˆ๋‹ค.

๋”ฅ๋Ÿฌ๋‹ ์ดˆ๋ณด์ž๋ถ€ํ„ฐ ์ „๋ฌธ๊ฐ€๊นŒ์ง€, ์ปดํ“จํ„ฐ ๋น„์ „ ํ”„๋กœ์ ํŠธ๋ฅผ ์œ„ํ•œ YOLO ์˜ ๊ตฌํ˜„๊ณผ ์ตœ์ ํ™”์— ๋Œ€ํ•œ ์œ ์šฉํ•œ ์ธ์‚ฌ์ดํŠธ๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ๋Š” ํŠœํ† ๋ฆฌ์–ผ์ž…๋‹ˆ๋‹ค. ์ง€๊ธˆ ๋ฐ”๋กœ ์‹œ์ž‘ํ•˜์„ธ์š”!



Watch: Ultralytics YOLOv8 ๊ฐ€์ด๋“œ ๊ฐœ์š”

๊ฐ€์ด๋“œ

๋‹ค์Œ์€ Ultralytics YOLO ์˜ ๋‹ค์–‘ํ•œ ์ธก๋ฉด์„ ์ตํžˆ๋Š” ๋ฐ ๋„์›€์ด ๋˜๋Š” ์‹ฌ์ธต ๊ฐ€์ด๋“œ ๋ชจ์Œ์ž…๋‹ˆ๋‹ค.

  • YOLO ์ผ๋ฐ˜์ ์ธ ๋ฌธ์ œ โญ ๊ถŒ์žฅ ์‚ฌํ•ญ: Ultralytics YOLO ๋ชจ๋ธ๋กœ ์ž‘์—…ํ•  ๋•Œ ๊ฐ€์žฅ ์ž์ฃผ ๋ฐœ์ƒํ•˜๋Š” ๋ฌธ์ œ์— ๋Œ€ํ•œ ์‹ค์šฉ์ ์ธ ์†”๋ฃจ์…˜ ๋ฐ ๋ฌธ์ œ ํ•ด๊ฒฐ ํŒ์ž…๋‹ˆ๋‹ค.
  • YOLO ์„ฑ๋Šฅ ์ง€ํ‘œ โญ ํ•„์ˆ˜: YOLO ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ๋˜๋Š” mAP, IoU ๋ฐ F1 ์ ์ˆ˜์™€ ๊ฐ™์€ ์ฃผ์š” ์ง€ํ‘œ๋ฅผ ์ดํ•ดํ•ฉ๋‹ˆ๋‹ค. ํƒ์ง€ ์ •ํ™•๋„์™€ ์†๋„๋ฅผ ๊ฐœ์„ ํ•˜๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ์‹ค์šฉ์ ์ธ ์˜ˆ์ œ์™€ ํŒ์ด ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.
  • ๋ชจ๋ธ ๋ฐฐํฌ ์˜ต์…˜: ONNX , OpenVINO, TensorRT ์™€ ๊ฐ™์€ YOLO ๋ชจ๋ธ ๋ฐฐํฌ ํ˜•์‹์— ๋Œ€ํ•œ ๊ฐœ์š”์™€ ๊ฐ๊ฐ์˜ ์žฅ๋‹จ์ ์„ ํ†ตํ•ด ๋ฐฐํฌ ์ „๋žต์— ๋Œ€ํ•œ ์ •๋ณด๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  • K-Fold ๊ต์ฐจ ๊ฒ€์ฆ ๐Ÿš€ ์‹ ๊ทœ: K-Fold ๊ต์ฐจ ๊ฒ€์ฆ ๊ธฐ๋ฒ•์„ ์‚ฌ์šฉํ•˜์—ฌ ๋ชจ๋ธ ์ผ๋ฐ˜ํ™”๋ฅผ ๊ฐœ์„ ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์•Œ์•„๋ณด์„ธ์š”.
  • ํ•˜์ดํผํŒŒ๋ผ๋ฏธํ„ฐ ํŠœ๋‹ ๐Ÿš€ ์‹ ๊ทœ: ํŠœ๋„ˆ ํด๋ž˜์Šค์™€ ์œ ์ „์  ์ง„ํ™” ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์‚ฌ์šฉํ•˜์—ฌ ํ•˜์ดํผํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ๋ฏธ์„ธ ์กฐ์ •ํ•˜์—ฌ YOLO ๋ชจ๋ธ์„ ์ตœ์ ํ™”ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์•Œ์•„๋ณด์„ธ์š”.
  • SAHI ํƒ€์ผํ˜• ์ถ”๋ก  ๐Ÿš€ ์‹ ๊ทœ: ๊ณ ํ•ด์ƒ๋„ ์ด๋ฏธ์ง€์—์„œ ๋ฌผ์ฒด ๊ฐ์ง€๋ฅผ ์œ„ํ•ด YOLOv8 ์—์„œ SAHI์˜ ์Šฌ๋ผ์ด์Šค ์ถ”๋ก  ๊ธฐ๋Šฅ์„ ํ™œ์šฉํ•˜๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ์ข…ํ•ฉ ๊ฐ€์ด๋“œ์ž…๋‹ˆ๋‹ค.
  • AzureML ๋น ๋ฅธ ์‹œ์ž‘ ๐Ÿš€ ์ƒˆ ๊ธฐ๋Šฅ: Microsoft์˜ Azure ๋จธ์‹  ๋Ÿฌ๋‹ ํ”Œ๋žซํผ์—์„œ Ultralytics YOLO ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜์—ฌ ์‹œ์ž‘ํ•˜๊ณ  ์‹คํ–‰ํ•˜์„ธ์š”. ํด๋ผ์šฐ๋“œ์—์„œ ๊ฐœ์ฒด ๊ฐ์ง€ ํ”„๋กœ์ ํŠธ๋ฅผ ํ›ˆ๋ จ, ๋ฐฐํฌ ๋ฐ ํ™•์žฅํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์•Œ์•„๋ณด์„ธ์š”.
  • Conda ๋น ๋ฅธ ์‹œ์ž‘ ๐Ÿš€ ์‹ ๊ทœ: Ultralytics ์— ๋Œ€ํ•œ Conda ํ™˜๊ฒฝ ์„ค์ •์— ๋Œ€ํ•œ ๋‹จ๊ณ„๋ณ„ ๊ฐ€์ด๋“œ์ž…๋‹ˆ๋‹ค. Ultralytics ํŒจํ‚ค์ง€๋ฅผ ํšจ์œจ์ ์œผ๋กœ ์„ค์น˜ํ•˜๊ณ  ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์•Œ์•„๋ณด์„ธ์š”.
  • Docker๋น ๋ฅธ ์‹œ์ž‘ ๐Ÿš€ ์‹ ๊ทœ: Docker์—์„œ Ultralytics YOLO ๋ชจ๋ธ์„ ์„ค์ •ํ•˜๊ณ  ์‚ฌ์šฉํ•˜๊ธฐ ์œ„ํ•œ ์ „์ฒด ๊ฐ€์ด๋“œ์ž…๋‹ˆ๋‹ค. ์ผ๊ด€๋œ ๊ฐœ๋ฐœ ๋ฐ ๋ฐฐํฌ๋ฅผ ์œ„ํ•ด Docker๋ฅผ ์„ค์น˜ํ•˜๊ณ , GPU ์ง€์›์„ ๊ด€๋ฆฌํ•˜๊ณ , ๊ฒฉ๋ฆฌ๋œ ์ปจํ…Œ์ด๋„ˆ์—์„œ YOLO ๋ชจ๋ธ์„ ์‹คํ–‰ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์•Œ์•„๋ณด์„ธ์š”.
  • ๋ผ์ฆˆ๋ฒ ๋ฆฌ ํŒŒ์ด ๐Ÿš€ ์‹ ๊ทœ: ์ตœ์‹  ๋ผ์ฆˆ๋ฒ ๋ฆฌ ํŒŒ์ด ํ•˜๋“œ์›จ์–ด์—์„œ YOLO ๋ชจ๋ธ์„ ์‹คํ–‰ํ•˜๋Š” ๋น ๋ฅธ ์‹œ์ž‘ ํŠœํ† ๋ฆฌ์–ผ.
  • NVIDIA-Jetson๐Ÿš€์‹ ๊ทœ: NVIDIA Jetson ์žฅ์น˜์— YOLO ๋ชจ๋ธ์„ ๋ฐฐํฌํ•˜๊ธฐ ์œ„ํ•œ ๋น ๋ฅธ ์‹œ์ž‘ ๊ฐ€์ด๋“œ.
  • Triton ์ถ”๋ก  ์„œ๋ฒ„ ํ†ตํ•ฉ ๐Ÿš€ ์‹ ๊ทœ: ํ™•์žฅ ๊ฐ€๋Šฅํ•˜๊ณ  ํšจ์œจ์ ์ธ ๋”ฅ ๋Ÿฌ๋‹ ์ถ”๋ก  ๋ฐฐํฌ๋ฅผ ์œ„ํ•ด Ultralytics YOLOv8 ๊ณผ NVIDIA์˜ Triton ์ถ”๋ก  ์„œ๋ฒ„์˜ ํ†ตํ•ฉ์— ๋Œ€ํ•ด ์ž์„ธํžˆ ์•Œ์•„๋ณด์„ธ์š”.
  • YOLO ์Šค๋ ˆ๋“œ ์•ˆ์ „ ์ถ”๋ก  ๐Ÿš€ ์‹ ๊ทœ: YOLO ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜์—ฌ ์Šค๋ ˆ๋“œ ์•ˆ์ „ ๋ฐฉ์‹์œผ๋กœ ์ถ”๋ก ์„ ์ˆ˜ํ–‰ํ•˜๊ธฐ ์œ„ํ•œ ๊ฐ€์ด๋“œ๋ผ์ธ์ž…๋‹ˆ๋‹ค. ๊ฒฝ์Ÿ ์กฐ๊ฑด์„ ๋ฐฉ์ง€ํ•˜๊ณ  ์ผ๊ด€๋œ ์˜ˆ์ธก์„ ๋ณด์žฅํ•˜๊ธฐ ์œ„ํ•œ ์Šค๋ ˆ๋“œ ์•ˆ์ „์˜ ์ค‘์š”์„ฑ๊ณผ ๋ชจ๋ฒ” ์‚ฌ๋ก€์— ๋Œ€ํ•ด ์•Œ์•„๋ณด์„ธ์š”.
  • ์„ธ๋ถ„ํ™” ๊ฐœ์ฒด ๋ถ„๋ฆฌํ•˜๊ธฐ ๐Ÿš€ ์‹ ๊ทœ: Ultralytics ์„ธ๋ถ„ํ™”๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ด๋ฏธ์ง€์—์„œ ๊ฐœ์ฒด๋ฅผ ์ถ”์ถœํ•˜๊ฑฐ๋‚˜ ๋ถ„๋ฆฌํ•˜๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ๋‹จ๊ณ„๋ณ„ ๋ ˆ์‹œํ”ผ์™€ ์„ค๋ช…์ด ์ถ”๊ฐ€๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
  • ๋ผ์ฆˆ๋ฒ ๋ฆฌ ํŒŒ์ด์˜ Edge TPU : ๊ตฌ๊ธ€ ์—ฃ์ง€ TPU, ๋ผ์ฆˆ๋ฒ ๋ฆฌ ํŒŒ์ด์—์„œ YOLO ์ถ”๋ก ์„ ๊ฐ€์†ํ™”ํ•ฉ๋‹ˆ๋‹ค.
  • ํ„ฐ๋ฏธ๋„์—์„œ ์ถ”๋ก  ์ด๋ฏธ์ง€ ๋ณด๊ธฐ: ์›๊ฒฉ ํ„ฐ๋„ ๋˜๋Š” SSH ์„ธ์…˜์„ ์‚ฌ์šฉํ•  ๋•Œ VSCode์˜ ํ†ตํ•ฉ ํ„ฐ๋ฏธ๋„์„ ์‚ฌ์šฉํ•˜์—ฌ ์ถ”๋ก  ๊ฒฐ๊ณผ๋ฅผ ๋ณผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  • OpenVINO ์ง€์—ฐ ์‹œ๊ฐ„ ๋Œ€ ์ฒ˜๋ฆฌ๋Ÿ‰ ๋ชจ๋“œ - ์ตœ๊ณ ์˜ YOLO ์ถ”๋ก  ์„ฑ๋Šฅ์„ ์œ„ํ•œ ์ง€์—ฐ ์‹œ๊ฐ„ ๋ฐ ์ฒ˜๋ฆฌ๋Ÿ‰ ์ตœ์ ํ™” ๊ธฐ๋ฒ•์— ๋Œ€ํ•ด ์•Œ์•„๋ณด์„ธ์š”.
  • Steps of a Computer Vision Project ๐Ÿš€ NEW: Learn about the key steps involved in a computer vision project, including defining goals, selecting models, preparing data, and evaluating results.
  • Defining A Computer Vision Project's Goals ๐Ÿš€ NEW: Walk through how to effectively define clear and measurable goals for your computer vision project. Learn the importance of a well-defined problem statement and how it creates a roadmap for your project.
  • Data Collection and Annotation๐Ÿš€ NEW: Explore the tools, techniques, and best practices for collecting and annotating data to create high-quality inputs for your computer vision models.
  • Preprocessing Annotated Data๐Ÿš€ NEW: Learn about preprocessing and augmenting image data in computer vision projects using YOLOv8, including normalization, dataset augmentation, splitting, and exploratory data analysis (EDA).

๊ฐ€์ด๋“œ์— ๊ธฐ์—ฌํ•˜๊ธฐ

์ปค๋ฎค๋‹ˆํ‹ฐ์˜ ๊ธฐ์—ฌ๋ฅผ ํ™˜์˜ํ•ฉ๋‹ˆ๋‹ค! ์•„์ง ๊ฐ€์ด๋“œ์—์„œ ๋‹ค๋ฃจ์ง€ ์•Š์€ Ultralytics YOLO ์˜ ํŠน์ • ์ธก๋ฉด์„ ๋งˆ์Šคํ„ฐํ–ˆ๋‹ค๋ฉด ์—ฌ๋Ÿฌ๋ถ„์˜ ์ „๋ฌธ ์ง€์‹์„ ๊ณต์œ ํ•ด ์ฃผ์‹œ๊ธฐ ๋ฐ”๋ž๋‹ˆ๋‹ค. ๊ฐ€์ด๋“œ๋ฅผ ์ž‘์„ฑํ•˜๋Š” ๊ฒƒ์€ ์ปค๋ฎค๋‹ˆํ‹ฐ์— ๋ณด๋‹ตํ•˜๊ณ  ๋”์šฑ ํฌ๊ด„์ ์ด๊ณ  ์‚ฌ์šฉ์ž ์นœํ™”์ ์ธ ๋ฌธ์„œ๋ฅผ ๋งŒ๋“œ๋Š” ๋ฐ ๋„์›€์ด ๋˜๋Š” ์ข‹์€ ๋ฐฉ๋ฒ•์ž…๋‹ˆ๋‹ค.

์‹œ์ž‘ํ•˜๋ ค๋ฉด ํ’€ ๋ฆฌํ€˜์ŠคํŠธ(PR)๋ฅผ ์—ฌ๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ๊ฐ€์ด๋“œ๋ผ์ธ์ด ๋‹ด๊ธด ๊ธฐ์—ฌ ๊ฐ€์ด๋“œ๋ฅผ ์ฐธ์กฐํ•˜์„ธ์š”(๐Ÿ› ๏ธ). ์—ฌ๋Ÿฌ๋ถ„์˜ ๊ธฐ์—ฌ๋ฅผ ๊ธฐ๋‹ค๋ฆฌ๊ฒ ์Šต๋‹ˆ๋‹ค!

Ultralytics YOLO ์ƒํƒœ๊ณ„๋ฅผ ๋”์šฑ ๊ฒฌ๊ณ ํ•˜๊ณ  ๋‹ค์–‘ํ•˜๊ฒŒ ๋งŒ๋“ค๊ธฐ ์œ„ํ•ด ํ•จ๊ป˜ ๋…ธ๋ ฅํ•ฉ์‹œ๋‹ค ๐Ÿ™!



Created 2023-11-12, Updated 2024-06-10
Authors: glenn-jocher (10), abirami-vina (2), RizwanMunawar (7), Burhan-Q (3), lakshanthad (1), ouphi (1)

๋Œ“๊ธ€