The 8th International Conference on Next Generation Computing 2022 (2022.10)바로가기
페이지
pp.205-206
저자
Hyojun Lee, Bowon Lee
언어
영어(ENG)
URL
https://www.earticle.net/Article/A419775
원문정보
초록
영어
In this paper, by cross-applying the DINO (DETR with Improved deNoising anchOrboxes) model to various datasets, we examine what characteristics of dataset are effective for traffic object detection training. DINO model is best DETR (DEtection with TRansformer)-like model in object detection. For the experiment, a total of two datasets were used: COCO and BDD100K datasets. As a result of evaluation with BDD100K dataset which contains diverse driving images, dataset with the same texture as the evaluation dataset showed similar performance with less data than the high texture dataset focused on each object.
목차
Abstract I. INTRODUCTION II. DATASET CROSS APPLICATION USING DINO MODEL A. DINO model B. Modify dataset C. Experiments and Results III. CONCLUSION REFERENCES