| 제목 |
영상의 연속성 기반 Key 개념 도입을 통해 인식 성능을 향상시킨
딥러닝 네트워크 개선 연구 |
| 분야 |
전기ㆍ전자ㆍ통신 |
| 언어 |
Korean |
| 저자 |
이원우(국민대학교), 최윤석(국민대학교), 유진우(국민대학교) |
| Key Words |
Deep leraning(딥러닝), Average precision(평균 정밀도), Recall(재현율), Bounding box(경계 상자), Loss function(손실 함수), Convolution(합성곱) |
| 초록 |
In this paper, we propose a novel deep learning algorithm to improve the detector’s performance in sequential images. By adopting a sliding window method that uses three consecutive images, the keys are generated by comparing the positions of the detected boxes for each of the images by using Generalized-IoU. Addition and Merge tasks were applied separately through comparison by using Complete-IoU after key generation. They have the effect of correcting the unexpected or incorrectly predicted bounding boxes. As a result, performance was improved in terms of average precision. |
| 원문(PDF) |
다운로드
Journal Site
|
