Yolo 활용 book segmentation 논문 방법론 정리

Research of YOLO Architecture Models in Book Detection

논문 링크 : https://www.researchgate.net/publication/346820617_Research_of_YOLO_Architecture_Models_in_BookDetection

Model

YOLO CONVOLUTIONAL NETWORK 활용
1. YOLOv3(3 outputs)
2. YOLOv3(2 outputs)
3. Tiny YOLOv3

data

• 500개 이미지(5,245권 책 포함)
  - augmentation : such as scaling, rotation, flip, cropping distortion of
  colours and so on.
• train 60%, valid 10%, test 30%

netword parameter

0. 800 epochs
1. size of input image – 448x448x3;
2. optimizer – Adam;
3. the learning rate – 0,0001;
4. batch-size – 2 samples;
5. metric for assessing the correctness of the class definition – AP.

Result