Wang Ao
Wang Ao
感谢您的关注! 1. 您好,导出的onnx默认是端到端的,会输出最终预测的box坐标,标签和置信度分数,因此没有80类中每个类的预测概率,具体输出格式您可以参考#7。 2. 您好,因为现在使用命令行时会根据文件名里是否有v10来判断是否用YOLOv10的模型,您可以把pt文件重命名为带yolov10名字的pt文件,或者您可以使用python api调用方式来进行predict,如 ``` from ultralytics import YOLOv10 model = YOLOv10("best.pt") model.predict(...) ```
No worries. ❤️
您好,感谢您的关注!您好,我们想确认一下您的训练环境是一致的吗?是都基于这个代码库训练YOLOv8和YOLOv10的吗?另一方面,您方便看下这个https://github.com/THU-MIG/yolov10/issues/111#issuecomment-2137277092 是否有帮助吗?谢谢!~
> > 或许应该都吧val结果保存下来**--save-json True **, 然后都用这个结果使用cocoapi进行验证比对 > > > > https://github.com/THU-MIG/yolov10/issues/100#issue-2320343475 > > https://github.com/THU-MIG/yolov10/issues/111#issuecomment-2138991593 > > > > 有可能是定制数据集里小目标比较多的原因吧??? 您好,关于小目标请您看一下#136,您的测试评估方式跟阈值有关吗?
Please feel free to reopen this issue if you have further questions.
Thanks for your interest! The iou argument is not needed for YOLOv10 and it will have no impact. It is meant to be compatible with other YOLOs.
Thanks for your interest! We just made the `YOLOv10` by default in cli. Could you please update the codebase and try again? Thank you!
@Hy87380510 Thanks. The FLOPs are correctly shown in our local environment. Is your code the latest version? 