megvii-research / MOTRv2

[CVPR2023] MOTRv2: Bootstrapping End-to-End Multi-Object Tracking by Pretrained Object Detectors
Other
343 stars 44 forks source link

About inference time? #60

Open dcstempt opened 6 months ago

dcstempt commented 6 months ago

Thank you for your nice work. Can this project achieve real-time video stream object tracking? During my testing, with the RTX A4000 GPU, the average inference time per frame is 110ms. When using the RTX 3050 GPU, the average inference time per frame is 200ms.

If I add a custom object detection head to YOLOx, the processing time will be longer.