Skip to content

Latest commit

ย 

History

History
131 lines (92 loc) ยท 5.71 KB

hustvl_yolop.md

File metadata and controls

131 lines (92 loc) ยท 5.71 KB
layout background-class body-class category title summary image author tags github-link github-id featured_image_1 featured_image_2 accelerator demo-model-link
hub_detail
hub-background
hub
researchers
YOLOP
YOLOP pretrained on the BDD100K dataset
yolop.png
Hust Visual Learning Team
vision
hustvl/YOLOP
no-image
no-image
cuda-optional

์‹œ์ž‘ํ•˜๊ธฐ ์ „ ์ฐธ๊ณ  ์‚ฌํ•ญ

YOLOP ์ข…์† ํŒจํ‚ค์ง€๋ฅผ ์„ค์น˜ํ•˜๋ ค๋ฉด ์•„๋ž˜ ๋ช…๋ น์„ ์ˆ˜ํ–‰ํ•ด์ฃผ์„ธ์š”:

pip install -qr https://github.com/hustvl/YOLOP/blob/main/requirements.txt  # install dependencies

YOLOP: You Only Look Once for Panoptic driving Perception

๋ชจ๋ธ ์„ค๋ช…

YOLOP Model

ย 
  • YOLOP๋Š” ์ž์œจ ์ฃผํ–‰์—์„œ ์ค‘์š”ํ•œ, ๋‹ค์Œ์˜ ์„ธ ๊ฐ€์ง€ ์ž‘์—…์„ ๊ณต๋™์œผ๋กœ ์ฒ˜๋ฆฌํ•  ์ˆ˜ ์žˆ๋Š” ํšจ์œจ์ ์ธ ๋‹ค์ค‘ ์ž‘์—… ๋„คํŠธ์›Œํฌ ์ž…๋‹ˆ๋‹ค. ์ด ๋‹ค์ค‘ ๋„คํŠธ์›Œํฌ๋Š” ๋ฌผ์ฒด ๊ฐ์ง€(object detection), ์ฃผํ–‰ ์˜์—ญ ๋ถ„ํ• (drivable area segmentation), ์ฐจ์„  ์ธ์‹(lane detection)์„ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ YOLOP๋Š” BDD100K ๋ฐ์ดํ„ฐ์…‹์—์„œ ์ตœ์‹  ๊ธฐ์ˆ (state-of-the-art)์˜ ์ˆ˜์ค€์„ ์œ ์ง€ํ•˜๋ฉด์„œ ์ž„๋ฒ ๋””๋“œ ๊ธฐ๊ธฐ์—์„œ ์‹ค์‹œ๊ฐ„์„ฑ์— ๋„๋‹ฌํ•œ ์ตœ์ดˆ์˜ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.

๊ฒฐ๊ณผ

์ฐจ๋Ÿ‰ ๊ฐ์ฒด(Traffic Object) ์ธ์‹ ๊ฒฐ๊ณผ

Model Recall(%) mAP50(%) Speed(fps)
Multinet 81.3 60.2 8.6
DLT-Net 89.4 68.4 9.3
Faster R-CNN 77.2 55.6 5.3
YOLOv5s 86.8 77.2 82
YOLOP(ours) 89.2 76.5 41

์ฃผํ–‰ ๊ฐ€๋Šฅ ์˜์—ญ ์ธ์‹ ๊ฒฐ๊ณผ

Model mIOU(%) Speed(fps)
Multinet 71.6 8.6
DLT-Net 71.3 9.3
PSPNet 89.6 11.1
YOLOP(ours) 91.5 41

์ฐจ์„  ์ธ์‹ ๊ฒฐ๊ณผ

Model mIOU(%) IOU(%)
ENet 34.12 14.64
SCNN 35.79 15.84
ENet-SAD 36.56 16.02
YOLOP(ours) 70.50 26.20

์กฐ๊ฑด ๋ณ€ํ™”์— ๋”ฐ๋ฅธ ๋ชจ๋ธ ํ‰๊ฐ€ 1 (Ablation Studies 1): End-to-end v.s. Step-by-step

Training_method Recall(%) AP(%) mIoU(%) Accuracy(%) IoU(%)
ES-W 87.0 75.3 90.4 66.8 26.2
ED-W 87.3 76.0 91.6 71.2 26.1
ES-D-W 87.0 75.1 91.7 68.6 27.0
ED-S-W 87.5 76.1 91.6 68.0 26.8
End-to-end 89.2 76.5 91.5 70.5 26.2

์กฐ๊ฑด ๋ณ€ํ™”์— ๋”ฐ๋ฅธ ๋ชจ๋ธ ํ‰๊ฐ€ 2 (Ablation Studies 2): Multi-task v.s. Single task

Training_method Recall(%) AP(%) mIoU(%) Accuracy(%) IoU(%) Speed(ms/frame)
Det(only) 88.2 76.9 - - - 15.7
Da-Seg(only) - - 92.0 - - 14.8
Ll-Seg(only) - - - 79.6 27.9 14.8
Multitask 89.2 76.5 91.5 70.5 26.2 24.4

์•ˆ๋‚ด:

  • ํ‘œ 4์—์„œ E, D, S, W๋Š” ์ธ์ฝ”๋”(Encoder), ๊ฒ€์ถœ ํ—ค๋“œ(Detect head), 2๊ฐœ์˜ ์„ธ๊ทธ๋จผํŠธ ํ—ค๋“œ(Segment heads) ์™€ ์ „์ฒด ๋„คํŠธ์›Œํฌ๋ฅผ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋ž˜์„œ ์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜(์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์ฒซ์งธ, ์ธ์ฝ”๋” ๋ฐ ๊ฒ€์ถœ ํ—ค๋“œ๋งŒ ํ•™์Šตํ•ฉ๋‹ˆ๋‹ค. ๊ทธ ํ›„, ์ธ์ฝ”๋” ๋ฐ ๊ฒ€์ถœ ํ—ค๋“œ๋ฅผ ๊ณ ์ •ํ•˜๊ณ  ๋‘ ๊ฐœ์˜ ๋ถ„ํ• (segmentation) ํ—ค๋“œ๋ฅผ ํ•™์Šตํ•ฉ๋‹ˆ๋‹ค. ๋งˆ์ง€๋ง‰์œผ๋กœ, ์ „์ฒด ๋„คํŠธ์›Œํฌ๋Š” ์„ธ ๊ฐ€์ง€ ์ž‘์—… ๋ชจ๋‘์— ๋Œ€ํ•ด ํ•จ๊ป˜ ํ•™์Šต๋ฉ๋‹ˆ๋‹ค.)์€ ED-S-W๋กœ ํ‘œ๊ธฐ๋˜๋ฉฐ, ๋‹ค๋ฅธ ์•Œ๊ณ ๋ฆฌ์ฆ˜๋„ ๋งˆ์ฐฌ๊ฐ€์ง€์ž…๋‹ˆ๋‹ค.

์‹œ๊ฐํ™”

์ฐจ๋Ÿ‰ ๊ฐ์ฒด(Traffic Object) ์ธ์‹ ๊ฒฐ๊ณผ

Traffic Object Detection Result

ย 

์ฃผํ–‰ ๊ฐ€๋Šฅ ์˜์—ญ ์ธ์‹ ๊ฒฐ๊ณผ

Drivable Area Segmentation Result

ย 

์ฐจ์„  ์ธ์‹ ๊ฒฐ๊ณผ

Lane Detection Result

ย 

์•ˆ๋‚ด:

  • ์ฐจ์„  ์ธ์‹์˜ ์‹œ๊ฐํ™” ๊ฒฐ๊ณผ๋Š” ์ด์ฐจํ•จ์ˆ˜ ํ˜•ํƒœ๋กœ ๊ทผ์‚ฌํ•˜๋Š” ๊ณผ์ •(quadratic fitting)์„ ํ†ตํ•ด ํ›„์ฒ˜๋ฆฌ(post processed) ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.

๋ฐฐํฌ

YOLOP ๋ชจ๋ธ์€ ์ด๋ฏธ์ง€ ์บก์ณ๋ฅผ Zed Camera๊ฐ€ ์žฅ์ฐฉ๋œ Jetson Tx2์—์„œ ์‹ค์‹œ๊ฐ„์œผ๋กœ ์ถ”๋ก ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์†๋„ ํ–ฅ์ƒ์„ ์œ„ํ•ด TensorRT๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ๋ชจ๋ธ์˜ ๋ฐฐํฌ์™€ ์ถ”๋ก ์„ ์œ„ํ•ด github code ์—์„œ ์ฝ”๋“œ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.

ํŒŒ์ดํ† ์น˜ ํ—ˆ๋ธŒ๋กœ๋ถ€ํ„ฐ ๋ชจ๋ธ ๋ถˆ๋Ÿฌ์˜ค๊ธฐ

์ด ์˜ˆ์ œ๋Š” ์‚ฌ์ „์— ํ•™์Šต๋œ YOLOP ๋ชจ๋ธ์„ ๋ถˆ๋Ÿฌ์˜ค๊ณ  ์ถ”๋ก ์„ ์œ„ํ•œ ์ด๋ฏธ์ง€๋ฅผ ๋ชจ๋ธ์— ์ „๋‹ฌํ•ฉ๋‹ˆ๋‹ค.

import torch

model = torch.hub.load('hustvl/yolop', 'yolop', pretrained=True)

img = torch.randn(1,3,640,640)
det_out, da_seg_out,ll_seg_out = model(img)

์ธ์šฉ(Citation)

See for more detail in github code and arxiv paper.

๋ณธ ๋…ผ๋ฌธ๊ณผ ์ฝ”๋“œ๊ฐ€ ์—ฌ๋Ÿฌ๋ถ„์˜ ์—ฐ๊ตฌ์— ์œ ์šฉํ•˜๋‹ค๊ณ  ํŒ๋‹จ๋˜๋ฉด, GitHub star๋ฅผ ์ฃผ๋Š” ๊ฒƒ๊ณผ ๋ณธ ๋…ผ๋ฌธ์„ ์ธ์šฉํ•˜๋Š” ๊ฒƒ์„ ๊ณ ๋ คํ•ด ์ฃผ์„ธ์š”: