Hello @jimmyli
I hope you are doing very well.
Q1.
In an internal team test, Using GStreamer + NNStreamer pipeline that can achieve 30 FPS with YOLOv5s model and width=640, height=480.
Q2.
The i.MX 8M Plus is primarily designed for inference workloads rather than training.
I hope this can helps to you.
Best regards,
Salas.