Inference yolov8s custom train model (getting low fps)

Device : Jeston Xavier NX (16 GB RAM), trt version:8.2.1.9, jetpack:4.6.1
converted yolov8s into engine(fp=16, imgsz=640) after train
at the time of inferencing on video and camera geting just 12-14 fps,
how can i imrove fps??
*Yolov4-tiny gives same fps on same device

Hi @uditbajpai0495,

this forum is specifically for cuOpt questions. Please visit https://forums.developer.nvidia.com/ and look for a forum that is aligned to Jetson Xavier and/or inferencing.