자율주행을 위한 다중작업학습에 관한 연구 = Study on Multi-Task Learning for Autonomous Driving

다국어 초록 (Multilingual Abstract)

For autonomous driving, we explored a method for safe autonomous driving based on the given hardware conditions, taking into account the performance (accuracy, processing speed) of image recognition tasks performed by the corresponding sensors. In particular, we analyzed the performance of multiple image recognition optimization tasks through multi-task learning (MTL), which can process several tasks simultaneously, and proposed a MDE (Multi-task Decision and Enhancement) algorithm for optimization. Using this MDE algorithm, it is possible to determine multiple working sets that can minimize the overall delay time while optimizing accuracy. As a result of the experiment, we achieved up to around 15-54% reduction in execution time with similar accuracy performance through this strategy.

참고문헌 (Reference)

1 곽대원 ; 유지상 ; 손민준 ; 박민수 ; 최동건 ; 이성진, "자율주행을 위한 실시간 차선인식 기술에 대한 고찰" 한국통신학회 48 (48): 589-599, 2023

2 배은지 ; 이성진, "이미지 분류 네트워크에서의 효율적 훈련 기법" 한국통신학회 46 (46): 1087-1096, 2021

3 J. Redmon, "You only look once : Unified, real-time object detection" 779-788, 2016

4 C. Han, "YOLOPv2: Better, faster, stronger for panoptic driving perception, arXiv preprint arXiv:2208.11434"

5 D. Wu, "YOLOP : You only look once for panoptic driving perception" 19 : 550-562, 2022

6 Z. Ge, "Y OLOX: Exceeding YOLO series in 2021, arX iv preprint arXiv:2107.08430"

7 Z. Qin, "Ultra fast structure aware deep lane detection" 1-14, 2022

8 O. Ronneberger, "U-Net : Convolutional networks for biomedical image segmentation" 9351 : 2015

9 X. Du, "SpineNet : Learning scale-permuted backbone for recognition and localization" 11589-11598, 2020

10 X. Lai, "Spherical transformer for LiDAR-based 3D recognition" 17545-17555, 2023

11 E. Xie, "SegFormer : Simple and efficient design for semantic segmentation with transformers" 34 : 12077-12090, 2021

12 A. Howard, "Searching for MobileNetV3" 1314-1324, 2019

13 W. Liu, "SSD : Single shot MultiBox detector" 2016

14 Z. Liu, "Rethinking the value of network pruning, arXiv preprint axXiv:1810. 05270"

15 D. Kwak, "Rethinking breast cancer diagnosis through deep learning based image recognition" 23 (23): 2307-, 2023

16 J. Guo, "Research on road scene understanding of autonomous vehicles based on multi-task learning" 23 (23): 2023

17 Y. J. Lee, "Recent R&D trends for lightweight deep learning" 34 (34): 40-50, 2019

18 A. Kirillov, "Panoptic feature pyramid networks" 2019

19 S. Shi, "PV-RCNN++ : Point-voxel feature set abstraction with local vector representation for 3d object detection" 131 : 531-551, 2023

20 Marvin Teichmann, "MultiNet: Real-time Joint Semantic Reasoning for Autonomous Driving" IEEE 1013-1020, 2018

21 K. Ishihara, "Multi-task learning with attention for end-to-end autonomous driving" 2896-2905, 2021

22 A. G. Howard, "MobileNets:Efficient convolutional neural networks for mobile vision applications, arXiv preprint arXiv:1704.04861"

23 M. Sandler, "MobileNetV2:Inverted residuals and linear bottlenecks" 4510-4520, 2018

24 W. Wang, "InternImage : Exploring largescale vision foundation models with deformable convolutions" 14408-14419, 2023

25 D. Vu, "HybridNets:End-to-end perception network, arXiv preprint arXiv:2203.09035"

26 T. Lin, "Focal loss for dense object detection" 42 (42): 318-327, 2020

27 T. Lin, "Feature pyramid networks for object detection" 936-944, 2017

28 D. -G. Lee, "Fast drivable areas estimation with multi-task learning for real-time autonomous driving assistant" 11 (11): 10713-, 2021

29 M. Tan, "EfficientDet : Scalable and efficient object detection" 10778-10787, 2020

30 A. I. Károly, "Deep learning in robotics : Survey on model structures and training strategies" 51 (51): 266-279, 2021

31 Y. Hong, "Crossmodality knowledge distillation network for monocular 3d object detection" 13670 : 2022

32 F. Yu, "BDD100K: A large-scale diverse driving video database"

33 S. Grigorescu, "A survey of deep learning techniques for autonomous driving" 37 (37): 2019

34 H. Lee, "A method of deep learning model optimization for image classification on edge device" 22 (22): 2022

35 J. Terven, "A comprehensive review of YOLO: From YOLOv1 to YOLOv8 and beyond, arXiv preprint arXiv:2304.00501"

36 Y. Kim, "3D Dual-Fusion: Dual-domain dualquery camera-lidar fusion for 3d object detection, arXiv Preprint arXiv:2211.13589"