Open CooperLuo32 opened 2 years ago
我们基于YOLOv5[51]算法6.0版本的L模型进行改进的,在输入端用Mosaic[50]数据增强,保留Slice[51]切片操作。 [50] Bochkovskiy A, Wang C Y, Liao H. YOLOv4: Optimal Speed and Accuracy of Object Detection[J]. arXiv:2004.10934, 2020. [51] ultralytics/yolov5: v6.0 [Online]. Available: https://zenodo.org/record/4418161#.YcQsLWBBxPY.
YOLOv5的模型超参和模型配置可以分享一下吗?复现用,谢谢了~~