์ฝ˜ํ…์ธ ๋กœ ๊ฑด๋„ˆ๋›ฐ๊ธฐ

์ด๋ฏธ์ง€๋„ท ๋ฐ์ดํ„ฐ ์„ธํŠธ

ImageNet์€ ์‹œ๊ฐ์  ๊ฐ์ฒด ์ธ์‹ ์—ฐ๊ตฌ์— ์‚ฌ์šฉํ•˜๋„๋ก ์„ค๊ณ„๋œ ์ฃผ์„์ด ๋‹ฌ๋ฆฐ ์ด๋ฏธ์ง€์˜ ๋Œ€๊ทœ๋ชจ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์ž…๋‹ˆ๋‹ค. 1,400๋งŒ ๊ฐœ ์ด์ƒ์˜ ์ด๋ฏธ์ง€๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ์œผ๋ฉฐ, ๊ฐ ์ด๋ฏธ์ง€์—๋Š” WordNet ๋™๊ธฐํ™” ์„ธํŠธ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ฃผ์„์„ ๋‹ฌ์•„ ์ปดํ“จํ„ฐ ๋น„์ „ ์ž‘์—…์—์„œ ๋”ฅ๋Ÿฌ๋‹ ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•˜๋Š” ๋ฐ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋Š” ๊ฐ€์žฅ ๊ด‘๋ฒ”์œ„ํ•œ ๋ฆฌ์†Œ์Šค ์ค‘ ํ•˜๋‚˜์ž…๋‹ˆ๋‹ค.

ImageNet ์‚ฌ์ „ ํ•™์Šต ๋ชจ๋ธ

๋ชจ๋ธ ํฌ๊ธฐ
(ํ”ฝ์…€)
acc
top1
ACC
TOP5
์†๋„
CPU ONNX
(ms)
์†๋„
A100 TensorRT
(ms)
๋งค๊ฐœ๋ณ€์ˆ˜
(M)
FLOPs
(B) at 640
YOLOv8n-cls 224 69.0 88.3 12.9 0.31 2.7 4.3
YOLOv8s-cls 224 73.8 91.7 23.4 0.35 6.4 13.5
YOLOv8m-cls 224 76.8 93.5 85.4 0.62 17.0 42.7
YOLOv8l-cls 224 76.8 93.5 163.0 0.87 37.5 99.7
YOLOv8x-cls 224 79.0 94.6 232.0 1.01 57.4 154.8

์ฃผ์š” ๊ธฐ๋Šฅ

  • ์ด๋ฏธ์ง€๋„ท์—๋Š” ์ˆ˜์ฒœ ๊ฐœ์˜ ๊ฐœ์ฒด ์นดํ…Œ๊ณ ๋ฆฌ๋ฅผ ์•„์šฐ๋ฅด๋Š” 1,400๋งŒ ๊ฐœ ์ด์ƒ์˜ ๊ณ ํ•ด์ƒ๋„ ์ด๋ฏธ์ง€๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.
  • ๋ฐ์ดํ„ฐ ์„ธํŠธ๋Š” ์›Œ๋“œ๋„ท ๊ณ„์ธต ๊ตฌ์กฐ์— ๋”ฐ๋ผ ๊ตฌ์„ฑ๋˜๋ฉฐ, ๊ฐ ๋™์˜์–ด๋Š” ์นดํ…Œ๊ณ ๋ฆฌ๋ฅผ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค.
  • ImageNet์€ ์ปดํ“จํ„ฐ ๋น„์ „ ๋ถ„์•ผ์˜ ๊ต์œก ๋ฐ ๋ฒค์น˜๋งˆํ‚น, ํŠนํžˆ ์ด๋ฏธ์ง€ ๋ถ„๋ฅ˜ ๋ฐ ๊ฐ์ฒด ๊ฐ์ง€ ์ž‘์—…์— ๋„๋ฆฌ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค.
  • ๋งค๋…„ ์—ด๋ฆฌ๋Š” ILSVRC(ImageNet ๋Œ€๊ทœ๋ชจ ์‹œ๊ฐ ์ธ์‹ ์ฑŒ๋ฆฐ์ง€)๋Š” ์ปดํ“จํ„ฐ ๋น„์ „ ์—ฐ๊ตฌ๋ฅผ ๋ฐœ์ „์‹œํ‚ค๋Š” ๋ฐ ์ค‘์š”ํ•œ ์—ญํ• ์„ ํ•ด์™”์Šต๋‹ˆ๋‹ค.

๋ฐ์ดํ„ฐ ์„ธํŠธ ๊ตฌ์กฐ

์ด๋ฏธ์ง€๋„ท ๋ฐ์ดํ„ฐ ์„ธํŠธ๋Š” ์›Œ๋“œ๋„ท ๊ณ„์ธต๊ตฌ์กฐ๋ฅผ ์‚ฌ์šฉํ•ด ๊ตฌ์„ฑ๋ฉ๋‹ˆ๋‹ค. ๊ณ„์ธต ๊ตฌ์กฐ์˜ ๊ฐ ๋…ธ๋“œ๋Š” ์นดํ…Œ๊ณ ๋ฆฌ๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ๊ฐ ์นดํ…Œ๊ณ ๋ฆฌ๋Š” ๋™์˜์–ด ๋ชจ์Œ์ธ ์‹ ์…‹(synset)์œผ๋กœ ์„ค๋ช…๋ฉ๋‹ˆ๋‹ค. ์ด๋ฏธ์ง€๋„ท์˜ ์ด๋ฏธ์ง€์—๋Š” ํ•˜๋‚˜ ์ด์ƒ์˜ ์‹ ์…‹์œผ๋กœ ์ฃผ์„์ด ์ถ”๊ฐ€๋˜์–ด ๋‹ค์–‘ํ•œ ๊ฐ์ฒด์™€ ๊ทธ ๊ด€๊ณ„๋ฅผ ์ธ์‹ํ•˜๊ธฐ ์œ„ํ•œ ํ›ˆ๋ จ ๋ชจ๋ธ์— ํ’๋ถ€ํ•œ ๋ฆฌ์†Œ์Šค๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.

์ด๋ฏธ์ง€๋„ท ๋Œ€๊ทœ๋ชจ ์‹œ๊ฐ ์ธ์‹ ์ฑŒ๋ฆฐ์ง€(ILSVRC)

๋งค๋…„ ์—ด๋ฆฌ๋Š” ImageNet ๋Œ€๊ทœ๋ชจ ์‹œ๊ฐ ์ธ์‹ ์ฑŒ๋ฆฐ์ง€(ILSVRC) ๋Š” ์ปดํ“จํ„ฐ ๋น„์ „ ๋ถ„์•ผ์—์„œ ์ค‘์š”ํ•œ ํ–‰์‚ฌ์ž…๋‹ˆ๋‹ค. ์ด ํ–‰์‚ฌ๋Š” ์—ฐ๊ตฌ์ž์™€ ๊ฐœ๋ฐœ์ž๊ฐ€ ํ‘œ์ค€ํ™”๋œ ํ‰๊ฐ€ ์ง€ํ‘œ๋ฅผ ํ†ตํ•ด ๋Œ€๊ทœ๋ชจ ๋ฐ์ดํ„ฐ ์„ธํŠธ์—์„œ ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ ๋ชจ๋ธ์„ ํ‰๊ฐ€ํ•  ์ˆ˜ ์žˆ๋Š” ํ”Œ๋žซํผ์„ ์ œ๊ณตํ•ด ์™”์Šต๋‹ˆ๋‹ค. ILSVRC๋Š” ์ด๋ฏธ์ง€ ๋ถ„๋ฅ˜, ๋ฌผ์ฒด ๊ฐ์ง€ ๋ฐ ๊ธฐํƒ€ ์ปดํ“จํ„ฐ ๋น„์ „ ์ž‘์—…์„ ์œ„ํ•œ ๋”ฅ๋Ÿฌ๋‹ ๋ชจ๋ธ ๊ฐœ๋ฐœ์—์„œ ์ƒ๋‹นํ•œ ๋ฐœ์ „์„ ์ด๋Œ์–ด๋ƒˆ์Šต๋‹ˆ๋‹ค.

์• ํ”Œ๋ฆฌ์ผ€์ด์…˜

์ด๋ฏธ์ง€๋„ท ๋ฐ์ดํ„ฐ ์„ธํŠธ๋Š” ์ด๋ฏธ์ง€ ๋ถ„๋ฅ˜, ๋ฌผ์ฒด ๊ฐ์ง€, ๋ฌผ์ฒด ์œ„์น˜ ํŒŒ์•… ๋“ฑ ๋‹ค์–‘ํ•œ ์ปดํ“จํ„ฐ ๋น„์ „ ์ž‘์—…์—์„œ ๋”ฅ๋Ÿฌ๋‹ ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•˜๊ณ  ํ‰๊ฐ€ํ•˜๋Š” ๋ฐ ๋„๋ฆฌ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค. AlexNet, VGG, ResNet๊ณผ ๊ฐ™์ด ๋„๋ฆฌ ์‚ฌ์šฉ๋˜๋Š” ๋”ฅ ๋Ÿฌ๋‹ ์•„ํ‚คํ…์ฒ˜๋Š” ImageNet ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐœ๋ฐœ๋˜๊ณ  ๋ฒค์น˜๋งˆํ‚น๋˜์—ˆ์Šต๋‹ˆ๋‹ค.

์‚ฌ์šฉ๋ฒ•

์ด๋ฏธ์ง€ ํฌ๊ธฐ๊ฐ€ 224x224์ธ ์ด๋ฏธ์ง€๋„ท ๋ฐ์ดํ„ฐ ์„ธํŠธ์—์„œ 100๊ฐœ์˜ ์—ํฌํฌ์— ๋Œ€ํ•œ ๋”ฅ๋Ÿฌ๋‹ ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•˜๋ ค๋ฉด ๋‹ค์Œ ์ฝ”๋“œ ์กฐ๊ฐ์„ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•œ ์ธ์ˆ˜์˜ ์ „์ฒด ๋ชฉ๋ก์€ ๋ชจ๋ธ ํ•™์Šต ํŽ˜์ด์ง€๋ฅผ ์ฐธ์กฐํ•˜์„ธ์š”.

์—ด์ฐจ ์˜ˆ์‹œ

from ultralytics import YOLO

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

# Train the model
results = model.train(data="imagenet", epochs=100, imgsz=224)
# Start training from a pretrained *.pt model
yolo train data=imagenet model=yolov8n-cls.pt epochs=100 imgsz=224

์ƒ˜ํ”Œ ์ด๋ฏธ์ง€ ๋ฐ ์ฃผ์„

ImageNet ๋ฐ์ดํ„ฐ ์„ธํŠธ์—๋Š” ์ˆ˜์ฒœ ๊ฐœ์˜ ๋ฌผ์ฒด ์นดํ…Œ๊ณ ๋ฆฌ๋ฅผ ์•„์šฐ๋ฅด๋Š” ๊ณ ํ•ด์ƒ๋„ ์ด๋ฏธ์ง€๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ์–ด ์ปดํ“จํ„ฐ ๋น„์ „ ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•˜๊ณ  ํ‰๊ฐ€ํ•  ์ˆ˜ ์žˆ๋Š” ๋‹ค์–‘ํ•˜๊ณ  ๊ด‘๋ฒ”์œ„ํ•œ ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์Œ์€ ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ๋ช‡ ๊ฐ€์ง€ ์ด๋ฏธ์ง€ ์˜ˆ์‹œ์ž…๋‹ˆ๋‹ค:

๋ฐ์ดํ„ฐ ์„ธํŠธ ์ƒ˜ํ”Œ ์ด๋ฏธ์ง€

์ด ์˜ˆ๋Š” ์ด๋ฏธ์ง€๋„ท ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ํฌํ•จ๋œ ์ด๋ฏธ์ง€์˜ ๋‹ค์–‘์„ฑ๊ณผ ๋ณต์žก์„ฑ์„ ๋ณด์—ฌ์ฃผ๋ฉฐ, ๊ฐ•๋ ฅํ•œ ์ปดํ“จํ„ฐ ๋น„์ „ ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•˜๋Š” ๋ฐ ์žˆ์–ด ๋‹ค์–‘ํ•œ ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ์ค‘์š”์„ฑ์„ ๊ฐ•์กฐํ•ฉ๋‹ˆ๋‹ค.

์ธ์šฉ ๋ฐ ๊ฐ์‚ฌ

์—ฐ๊ตฌ ๋˜๋Š” ๊ฐœ๋ฐœ ์ž‘์—…์— ImageNet ๋ฐ์ดํ„ฐ์…‹์„ ์‚ฌ์šฉํ•˜๋Š” ๊ฒฝ์šฐ ๋‹ค์Œ ๋…ผ๋ฌธ์„ ์ธ์šฉํ•ด ์ฃผ์„ธ์š”:

@article{ILSVRC15,
         author = {Olga Russakovsky and Jia Deng and Hao Su and Jonathan Krause and Sanjeev Satheesh and Sean Ma and Zhiheng Huang and Andrej Karpathy and Aditya Khosla and Michael Bernstein and Alexander C. Berg and Li Fei-Fei},
         title={ImageNet Large Scale Visual Recognition Challenge},
         year={2015},
         journal={International Journal of Computer Vision (IJCV)},
         volume={115},
         number={3},
         pages={211-252}
}

๋จธ์‹  ๋Ÿฌ๋‹ ๋ฐ ์ปดํ“จํ„ฐ ๋น„์ „ ์—ฐ๊ตฌ ์ปค๋ฎค๋‹ˆํ‹ฐ๋ฅผ ์œ„ํ•œ ๊ท€์ค‘ํ•œ ๋ฆฌ์†Œ์Šค์ธ ImageNet ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ๋งŒ๋“ค๊ณ  ์œ ์ง€ ๊ด€๋ฆฌํ•ด ์ฃผ์‹  Olga Russakovsky, Jia Deng, Li Fei-Fei๊ฐ€ ์ด๋„๋Š” ImageNet ํŒ€์— ๊ฐ์‚ฌ์˜ ๋ง์”€์„ ์ „ํ•ฉ๋‹ˆ๋‹ค. ImageNet ๋ฐ์ดํ„ฐ ์„ธํŠธ์™€ ์ œ์ž‘์ž์— ๋Œ€ํ•œ ์ž์„ธํ•œ ๋‚ด์šฉ์€ ImageNet ์›น์‚ฌ์ดํŠธ๋ฅผ ์ฐธ์กฐํ•˜์„ธ์š”.



Created 2023-11-12, Updated 2024-06-02
Authors: glenn-jocher (7), RizwanMunawar (1)

๋Œ“๊ธ€