YOLO-P1P2-CBAM: IMPROVING YOLO11 SOCCER BALL DETECTION USING EXTRA HIGH-RESOLUTION HEADS AND CBAM
Annotatsiya
Reliable soccer-ball detection is a key building block for sports analytics, broadcast enhancement, and robotic training systems, yet it remains challenging because the ball often occupies a very small region and appears under motion blur, occlusion, and complex backgrounds. This paper presents YOLO-P1P2-CBAM, a lightweight modification of an Ultralytics YOLO11 detector that (i) adds extra high-resolution detection heads (P1 and P2) to better preserve fine spatial details for tiny targets and (ii) injects Convolutional Block Attention Modules (CBAM) to refine multi-scale features before prediction. Experiments on a single-class soccer-ball dataset (2,474 images; 1,978/246/250 train/val/test) show that YOLO-P1P2-CBAM achieves 0.9398 mAP@0.5 and 0.6025 mAP@0.5:0.95, outperforming YOLO11n/s/m/l/x and two additional baselines trained under identical settings. Qualitative results indicate improved localization stability in cluttered scenes and at long ranges.
Kalit soʻzlar:
Iqtiboslar
S. Woo, J. Park, J.-Y. Lee, and I. S. Kweon, “CBAM: Convolutional Block Attention Module,” in Proc. ECCV, 2018.
Ultralytics, “YOLO11 Documentation,” 2024. [Online]. Available: https://docs.ultralytics.com/models/yolo11/
Ultralytics, “ultralytics/ultralytics (GitHub repository),” 2024. [Online]. Available: https://github.com/ultralytics/ultralytics
J. Komorowski, G. Kurzejamski, and G. Sarwas, “DeepBall: Deep Neural-Network Ball Detector,” arXiv:1902.07304, 2019.
D. Barry et al., “xYOLO: A Model for Real-Time Object Detection in Humanoid Soccer on Low-End Hardware,” arXiv:1910.03159, 2019.
Oʻquvchilar geografiyasi
Yuklab olishlar
Nashr qilingan
Son
Boʻlim
Iqtibos keltirish tartibi
