Główne pojęcia
This research paper introduces CST-YOLO, a novel object detection model specifically designed for small-scale objects like blood cells. By integrating a CNN-Swin Transformer module with an enhanced YOLOv7 architecture, CST-YOLO achieves superior detection accuracy compared to existing YOLO models and demonstrates the potential of CNN-Transformer fusion for improving small object detection.
Statystyki
CST-YOLO achieves 92.7%, 95.6%, and 91.1% mAP@0.5 on the BCCD, CBC, and BCD datasets, respectively.
YOLOv7 achieves 89.6%, 94.1%, and 87.8% mAP@0.5 on the BCCD, CBC, and BCD datasets, respectively.
YOLOv5x achieves 92.3%, 95.5%, and 88.4% mAP@0.5 on the BCCD, CBC, and BCD datasets, respectively.
CST-YOLO has 47.5M parameters.
YOLOv7 has 36.9M parameters.
YOLOv5x has 86.7M parameters.