increase FPS #2743
Replies: 3 comments 3 replies
-
@hassanSayedd for CPU-only deployment destinations like Pi 4 the best option at the moment may be ONNX export with ONNX runtime (you may see a 3-4x speedup according to some sources), or also pruning and quantization aware training (QAT), which we don't fully support yet but may in the future. |
Beta Was this translation helpful? Give feedback.
-
@hassanSayedd of course, a second option altogether is to train a smaller architecture, ie a YOLOv5xs size, by reducing the compound scaling constants in the YOLOv5s model and then training again. You'd have to experiment with the best tradeoff between reducing the two scaling constants here: Lines 1 to 5 in c03d590 |
Beta Was this translation helpful? Give feedback.
-
I'm getting 15fps on a jetson nano with.. |
Beta Was this translation helpful? Give feedback.
-
How can i increase FPS to be capable with low specification like raspberry pi 4 ?!
i trained YOLO5s but it has low FPS
and if want to deal with only the return value from the model and i do not want th show it how can i do that
thanks in advance
Beta Was this translation helpful? Give feedback.
All reactions