[DL] CCTV 차종 분류 모델_1

Dataset

• Dataset size: 3000 images
• Each class: 500 images
• Train data(2100개) : Test data(900개) = 7 : 3
• Classes
  • Car, Threewheel, Bus, Truck, Motorbike, Van
• 출처: kaggle

YOLOv5

CASE 1

• image size: 500
• batch size: 4
• epochs: 15
• weights: yolov5s.pt

• precision: 0.954, recall: 0.929, mAP50: 0.979, mAP50-95: 0.847

CASE 2

• image size: 640
• batch size: 8
• epochs: 50
• weights: yolov5s.pt

• precision: 0.956, recall: 0.935, mAP50: 0.974, mAP50-95: 0.869

CASE 3

• image size: 416
• batch size: 16
• epochs: 50
• weights: yolov5s.pt

• precision: 0.982, recall: 0.946, mAP50: 0.981, mAP50-95: 0.887

YOLOv7

CASE 4

• image size: 500
• batch size: 4
• epochs: 15
• weights: yolov7.pt

• precision: 0.975, recall: 0.943, mAP50: 0.981, mAP50-95: 0.871

모델 테스트(CASE 2)

• 원본 이미지 출처: https://pixabay.com/ko/