v8 vs nas about yolo-nas-onnx HOT 2 CLOSED

Hello, from my perspective I think the best model to deploy right at the edge on the web is the smaller one. Because you don't want the user get bored watching the loading screen to wait the model to be loaded. YOLOv8 give a smaller model size, yolov8n the smallest model which is around 13 MB if you compare it with the smallest YOLO-NAS model yolo_nas_s have the size around 48 MB. The smallest YOLO-NAS model is in par with yolov8s with 43 MB size.

For inferencing I haven't do benchmark yet since benchmarking model in the web is a pain 😀 .

But if you take a look of the performance YOLO-NAS claimed they have better MAP than YOLOv8 shown in the image bellow,

And the official YOLOv8 repo hasn't update the model yet so I expect YOLO-NAS have a better detection.

I haven't tried quantization on YOLO-NAS model but I expect the size doesn't drop to the level of yolov8n model.

from yolo-nas-onnx.

Thanks! I appreciate your thorough response.

from yolo-nas-onnx.