dcstempt

你应该在YOLOx的issue下面留言。ByteTrack本身不涉及任何深度学习的模型，他只是一个匹配预测算法而已。

我用A4000显卡，采用C++/TensorRT推理加速，YOLOv8s模型为检测头，模型推理一帧6ms，跟踪预测一帧，大概在500微秒的样子。总共一帧的预测+跟踪，大概就是6.5ms

Also this is my question. In author's work and paper, the YOLOx's proposal only support bounding box and score, but do not include target's class information.

the right key is "'DanceTrack\test\dancetrack0003\img1\00000001.txt'", not "'DanceTrack/test\dancetrack0003\img1\00000001.txt'". Do not use "/"