Open yuexiemail opened 2 years ago
❔Question
Hi, thank you your code. I use your code to process the UAVDT data. but the AP on UAVDT with Yolov5 is more than 99%, as shown in below:
It is unbelievable becuase the AP on UAVDT of existing detectors in paper is about 40%.
Do you encounter the problem?
Additional context
@yuexiemail hi, the mAP@50 in my experiments is around 77%, trained by YOLOv5m. 99% is unrealistic, try different size of YOLOv5 and see what is the difference
i have same issue, check box is right
I has compared with papers based on UAVDT, their used matlab toolkit to generate results. May be this is the reason of issue.
Pls did your problem solve, I was training with yolov8 recently and had the same problem
❔Question
Hi, thank you your code. I use your code to process the UAVDT data. but the AP on UAVDT with Yolov5 is more than 99%, as shown in below:
It is unbelievable becuase the AP on UAVDT of existing detectors in paper is about 40%.
Do you encounter the problem?
Additional context