์ฝ˜ํ…์ธ ๋กœ ๊ฑด๋„ˆ๋›ฐ๊ธฐ

Roboflow ์œ ๋‹ˆ๋ฒ„์Šค ์นดํŒŒ์ธ  ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ์„ธํŠธ

์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ Roboflow ์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ๋Š” ์ปดํ“จํ„ฐ ๋น„์ „ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์„ ์œ„ํ•ด ์„ค๊ณ„๋œ ์ด๋ฏธ์ง€์™€ ๋น„๋””์˜ค์˜ ์„ ๋ณ„๋œ ์ปฌ๋ ‰์…˜์œผ๋กœ, ํŠนํžˆ ์ž๋™์ฐจ ๋ถ€ํ’ˆ๊ณผ ๊ด€๋ จ๋œ ์„ธ๋ถ„ํ™” ์ž‘์—…์— ์ค‘์ ์„ ๋‘๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ ์„ธํŠธ๋Š” ์—ฌ๋Ÿฌ ๊ด€์ ์—์„œ ์บก์ฒ˜ํ•œ ๋‹ค์–‘ํ•œ ์‹œ๊ฐ์  ์ž๋ฃŒ๋ฅผ ์ œ๊ณตํ•˜๋ฉฐ, ์„ธ๋ถ„ํ™” ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•˜๊ณ  ํ…Œ์ŠคํŠธํ•˜๋Š” ๋ฐ ์œ ์šฉํ•œ ์ฃผ์„์ด ๋‹ฌ๋ฆฐ ์˜ˆ์ œ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.

์ž๋™์ฐจ ์—ฐ๊ตฌ, ์ฐจ๋Ÿ‰ ์œ ์ง€๋ณด์ˆ˜๋ฅผ ์œ„ํ•œ AI ์†”๋ฃจ์…˜ ๊ฐœ๋ฐœ, ์ปดํ“จํ„ฐ ๋น„์ „ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜ ํƒ์ƒ‰ ๋“ฑ ์–ด๋–ค ๋ถ„์•ผ์—์„œ๋“  ์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ๋Š” ํ”„๋กœ์ ํŠธ์˜ ์ •ํ™•์„ฑ๊ณผ ํšจ์œจ์„ฑ์„ ๋†’์ด๋Š” ๋ฐ ์œ ์šฉํ•œ ๋ฆฌ์†Œ์Šค๋กœ ํ™œ์šฉ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

๋ฐ์ดํ„ฐ ์„ธํŠธ ๊ตฌ์กฐ

์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ ๋‚ด์˜ ๋ฐ์ดํ„ฐ ๋ถ„ํฌ๋Š” ์•„๋ž˜์™€ ๊ฐ™์ด ๊ตฌ์„ฑ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค:

  • ํŠธ๋ ˆ์ด๋‹ ์„ธํŠธ: 3156๊ฐœ์˜ ์ด๋ฏธ์ง€์™€ ๊ฐ ์ด๋ฏธ์ง€์— ํ•ด๋‹นํ•˜๋Š” ์ฃผ์„์ด ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.
  • ํ…Œ์ŠคํŠธ ์„ธํŠธ: 276๊ฐœ์˜ ์ด๋ฏธ์ง€๋กœ ๊ตฌ์„ฑ๋˜๋ฉฐ, ๊ฐ ์ด๋ฏธ์ง€๋Š” ๊ฐ๊ฐ์˜ ์ฃผ์„๊ณผ ์ง์„ ์ด๋ฃน๋‹ˆ๋‹ค.
  • ์œ ํšจ์„ฑ ๊ฒ€์‚ฌ ์„ธํŠธ: 401๊ฐœ์˜ ์ด๋ฏธ์ง€๋กœ ๊ตฌ์„ฑ๋˜๋ฉฐ, ๊ฐ ์ด๋ฏธ์ง€์—๋Š” ํ•ด๋‹น ์ฃผ์„์ด ์žˆ์Šต๋‹ˆ๋‹ค.

์• ํ”Œ๋ฆฌ์ผ€์ด์…˜

์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™”๋Š” ์ž๋™์ฐจ ํ’ˆ์งˆ ๊ด€๋ฆฌ, ์ž๋™์ฐจ ์ˆ˜๋ฆฌ, ์ „์ž์ƒ๊ฑฐ๋ž˜ ์นดํƒˆ๋กœ๊ทธ, ๊ตํ†ต ๋ชจ๋‹ˆํ„ฐ๋ง, ์ž์œจ์ฃผํ–‰์ฐจ, ๋ณดํ—˜ ์ฒ˜๋ฆฌ, ์žฌํ™œ์šฉ, ์Šค๋งˆํŠธ ์‹œํ‹ฐ ์ด๋‹ˆ์…”ํ‹ฐ๋ธŒ ๋“ฑ์— ํ™œ์šฉ๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ์ฐจ๋Ÿ‰ ๋ถ€ํ’ˆ์„ ์ •ํ™•ํ•˜๊ฒŒ ์‹๋ณ„ํ•˜๊ณ  ๋ถ„๋ฅ˜ํ•˜์—ฌ ํ”„๋กœ์„ธ์Šค๋ฅผ ๊ฐ„์†Œํ™”ํ•จ์œผ๋กœ์จ ๋‹ค์–‘ํ•œ ์‚ฐ์—…์—์„œ ํšจ์œจ์„ฑ๊ณผ ์ž๋™ํ™”์— ๊ธฐ์—ฌํ•ฉ๋‹ˆ๋‹ค.

๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ YAML

๋ฐ์ดํ„ฐ ์„ธํŠธ ๊ตฌ์„ฑ์„ ์ •์˜ํ•˜๋Š” ๋ฐ๋Š” YAML(๋˜ ๋‹ค๋ฅธ ๋งˆํฌ์—… ์–ธ์–ด) ํŒŒ์ผ์ด ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—๋Š” ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ๊ฒฝ๋กœ, ํด๋ž˜์Šค ๋ฐ ๊ธฐํƒ€ ๊ด€๋ จ ์ •๋ณด๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ํŒจํ‚ค์ง€ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ๊ฒฝ์šฐ, ํŒจํ‚ค์ง€ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ carparts-seg.yaml ํŒŒ์ผ์€ ๋‹ค์Œ ์œ„์น˜์—์„œ ๊ด€๋ฆฌ๋ฉ๋‹ˆ๋‹ค. https://github.com/ultralytics/ultralytics/blob/main/ultralytics/cfg/datasets/carparts-seg.yaml.

ultralytics/cfg/datasets/carparts-seg.yaml

# Ultralytics YOLO ๐Ÿš€, AGPL-3.0 license
# Carparts-seg dataset by Ultralytics
# Documentation: https://docs.ultralytics.com/datasets/segment/carparts-seg/
# Example usage: yolo train data=carparts-seg.yaml
# parent
# โ”œโ”€โ”€ ultralytics
# โ””โ”€โ”€ datasets
#     โ””โ”€โ”€ carparts-seg  โ† downloads here (132 MB)

# Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]
path: ../datasets/carparts-seg # dataset root dir
train: train/images # train images (relative to 'path') 3516 images
val: valid/images # val images (relative to 'path') 276 images
test: test/images # test images (relative to 'path') 401 images

# Classes
names:
  0: back_bumper
  1: back_door
  2: back_glass
  3: back_left_door
  4: back_left_light
  5: back_light
  6: back_right_door
  7: back_right_light
  8: front_bumper
  9: front_door
  10: front_glass
  11: front_left_door
  12: front_left_light
  13: front_light
  14: front_right_door
  15: front_right_light
  16: hood
  17: left_mirror
  18: object
  19: right_mirror
  20: tailgate
  21: trunk
  22: wheel

# Download script/URL (optional)
download: https://ultralytics.com/assets/carparts-seg.zip

์‚ฌ์šฉ๋ฒ•

์ด๋ฏธ์ง€ ํฌ๊ธฐ๊ฐ€ 640์ธ 100๊ฐœ์˜ ์—ํฌํฌ์— ๋Œ€ํ•œ ์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ๋Œ€ํ•ด Ultralytics YOLOv8n ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•˜๋ ค๋ฉด ๋‹ค์Œ ์ฝ”๋“œ ์กฐ๊ฐ์„ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•œ ์ธ์ˆ˜์˜ ์ „์ฒด ๋ชฉ๋ก์€ ๋ชจ๋ธ ํ›ˆ๋ จ ํŽ˜์ด์ง€๋ฅผ ์ฐธ์กฐํ•˜์„ธ์š”.

์—ด์ฐจ ์˜ˆ์‹œ

from ultralytics import YOLO

# Load a model
model = YOLO('yolov8n-seg.pt')  # load a pretrained model (recommended for training)

# Train the model
results = model.train(data='carparts-seg.yaml', epochs=100, imgsz=640)
# Start training from a pretrained *.pt model
yolo segment train data=carparts-seg.yaml model=yolov8n-seg.pt epochs=100 imgsz=640

์ƒ˜ํ”Œ ๋ฐ์ดํ„ฐ ๋ฐ ์ฃผ์„

์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ์—๋Š” ๋‹ค์–‘ํ•œ ๊ด€์ ์—์„œ ์ดฌ์˜๋œ ๋‹ค์–‘ํ•œ ์ด๋ฏธ์ง€์™€ ๋™์˜์ƒ์ด ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ์•„๋ž˜์—์„œ ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ๋ฐ์ดํ„ฐ ์˜ˆ์‹œ์™€ ํ•ด๋‹น ์ฃผ์„์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค:

๋ฐ์ดํ„ฐ ์„ธํŠธ ์ƒ˜ํ”Œ ์ด๋ฏธ์ง€

  • ์ด ์ด๋ฏธ์ง€๋Š” ์ƒ˜ํ”Œ ๋‚ด์˜ ๊ฐ์ฒด ์„ธ๋ถ„ํ™”๋ฅผ ๋ณด์—ฌ์ฃผ๋ฉฐ, ์‹๋ณ„๋œ ๊ฐ์ฒด๋ฅผ ๋‘˜๋Ÿฌ์‹ผ ๋งˆ์Šคํฌ๊ฐ€ ์žˆ๋Š” ์ฃผ์„์ด ๋‹ฌ๋ฆฐ ๊ฒฝ๊ณ„ ์ƒ์ž๊ฐ€ ํŠน์ง•์ž…๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ ์„ธํŠธ๋Š” ๋‹ค์–‘ํ•œ ์œ„์น˜, ํ™˜๊ฒฝ, ๋ฐ€๋„์—์„œ ์บก์ฒ˜ํ•œ ๋‹ค์–‘ํ•œ ์ด๋ฏธ์ง€ ์„ธํŠธ๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์œผ๋ฉฐ, ์ด ์ž‘์—…์— ํŠนํ™”๋œ ๋ชจ๋ธ์„ ์ œ์ž‘ํ•˜๋Š” ๋ฐ ํ•„์š”ํ•œ ํฌ๊ด„์ ์ธ ๋ฆฌ์†Œ์Šค ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค.
  • ์ด ์‚ฌ๋ก€๋Š” ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ๋‚ด์žฌ๋œ ๋‹ค์–‘์„ฑ๊ณผ ๋ณต์žก์„ฑ์„ ๊ฐ•์กฐํ•˜๋ฉฐ ์ปดํ“จํ„ฐ ๋น„์ „ ์ž‘์—…, ํŠนํžˆ ์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ์˜์—ญ์—์„œ ๊ณ ํ’ˆ์งˆ ๋ฐ์ดํ„ฐ์˜ ์ค‘์š”ํ•œ ์—ญํ• ์„ ๊ฐ•์กฐํ•ฉ๋‹ˆ๋‹ค.

์ธ์šฉ ๋ฐ ๊ฐ์‚ฌ

์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ์—ฐ๊ตฌ ๋˜๋Š” ๊ฐœ๋ฐœ ํ”„๋กœ์ ํŠธ์— ํ†ตํ•ฉํ•˜๋Š” ๊ฒฝ์šฐ ๋‹ค์Œ ๋ฐฑ์„œ๋ฅผ ์ฐธ์กฐํ•˜์„ธ์š”:

   @misc{ car-seg-un1pm_dataset,
        title = { car-seg Dataset },
        type = { Open Source Dataset },
        author = { Gianmarco Russo },
        howpublished = { \url{ https://universe.roboflow.com/gianmarco-russo-vt9xr/car-seg-un1pm } },
        url = { https://universe.roboflow.com/gianmarco-russo-vt9xr/car-seg-un1pm },
        journal = { Roboflow Universe },
        publisher = { Roboflow },
        year = { 2023 },
        month = { nov },
        note = { visited on 2024-01-24 },
    }

์ฐจ๋Ÿ‰ ์œ ์ง€๋ณด์ˆ˜ ๋ฐ ์—ฐ๊ตฌ ํ”„๋กœ์ ํŠธ๋ฅผ ์œ„ํ•œ ๊ท€์ค‘ํ•œ ๋ฆฌ์†Œ์Šค์ธ ์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ๊ฐœ๋ฐœํ•˜๊ณ  ๊ด€๋ฆฌํ•˜๋Š” ๋ฐ ํ—Œ์‹ ํ•ด ์ฃผ์‹  Roboflow ํŒ€์—๊ฒŒ ๊ฐ์‚ฌ์˜ ๋ง์”€์„ ์ „ํ•ฉ๋‹ˆ๋‹ค. ์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ์™€ ์ œ์ž‘์ž์— ๋Œ€ํ•œ ์ž์„ธํ•œ ๋‚ด์šฉ์€ ์ž๋™์ฐจ ๋ถ€ํ’ˆ ์„ธ๋ถ„ํ™” ๋ฐ์ดํ„ฐ ์„ธํŠธ ํŽ˜์ด์ง€๋ฅผ ์ฐธ์กฐํ•˜์„ธ์š”.



์ƒ์„ฑ 2024-01-25, ์—…๋ฐ์ดํŠธ 2024-02-08
์ž‘์„ฑ์ž: chr043416@gmail.com (1), glenn-jocher (1)

๋Œ“๊ธ€