상세정보
학술대회
A Study on the Application of Knowledge Distillation in Ship Type Classification Model Development
- 발행일
- 202110
- 출처
- International Conference on Information and Communication Technology Convergence (ICTC) 2021, pp.280-282
- 협약과제
- 21HH5800, 영상 내 객체간 관계 분석 기반 해상 선박/구조물 상세 식별 콘텐츠 기술 개발, 남도원
- 초록
- In order to identify an object of interest in a given image with high accuracy, it is essential to select cumbersome deep learning models. However, these models are computationally expansive and cannot be deployed when limited environments. We experiment on a ship type classification model that supports fast inference while inheriting the performance of the teacher network through knowledge distillation and share the results. As can be seen from the experimental results, through knowledge distillation, we got a clue that we can learn a deep learning network that can provide real services. Through additional data learning and hyperparameter tuning, it can be deployed in the actual ship type classification system and utilized in various applications.
- KSP 제안 키워드
- Classification models, Classification system, Data Learning, Deep learning network, High accuracy, Ship type classification, deep learning(DL), deep learning models, knowledge distillation, model development