상세정보
학술대회
Action-Decision Networks for Visual Tracking with Deep Reinforcement Learning
- 저자
- 윤상두, 최종원, 유영준, 윤기민, 최진영
- 발행일
- 201707
- 출처
- Conference on Computer Vision and Pattern Recognition (CVPR) 2017, pp.1349-1358
- 협약과제
- 17HS3600, (1세부) 실시간 대규모 영상 데이터 이해·예측을 위한 고성능 비주얼 디스커버리 플랫폼 개발, 박종열
- 초록
- This paper proposes a novel tracker which is controlled by sequentially pursuing actions learned by deep reinforcement learning. In contrast to the existing trackers using deep networks, the proposed tracker is designed to achieve a light computation as well as satisfactory tracking accuracy in both location and scale. The deep network to control actions is pre-trained using various training sequences and fine-tuned during tracking for online adaptation to target and background changes. The pre-training is done by utilizing deep reinforcement learning as well as supervised learning. The use of reinforcement learning enables even partially labeled data to be successfully utilized for semi-supervised learning. Through evaluation of the OTB dataset, the proposed tracker is validated to achieve a competitive performance that is three times faster than state-of-the-art, deep network-based trackers. The fast version of the proposed method, which operates in real-time on GPU, outperforms the state-of-the-art real-time trackers.
- KSP 제안 키워드
- Competitive performance, Decision networks, Deep reinforcement learning, Online adaptation, Partially labeled data, Pre-Training, Real-Time, Reinforcement Learning(RL), Semi-Supervised Learning(SSL), Target and background, Visual Tracking