model-benchmark

this repo is a simple code for evaluating pytorch timm models (params, flops, fps) this repo used ptflops

bechmark

CPU: Intel(R) Core(TM) i5-1038NG7 CPU @ 2.00GHz CPU and no GPU.

Model	Resolution	Params(M)	GFLOPs	FPS
convnext_nano	(3, 224, 224)	15.59	2.46	16.8
convnext_nano_hnf	(3, 224, 224)	15.59	2.46	16.9
convnext_nano_ols	(3, 224, 224)	15.61	2.51	16.2

example

python3 eval.py --device cpu --iter 20

Model                          Resolution       Params     GFLOPs      FPS       
convnext_nano                  [(3, 224, 224), '15.59 M', '2.46 GMac', 16.8]
convnext_nano_hnf              [(3, 224, 224), '15.59 M', '2.46 GMac', 16.9]
convnext_nano_ols              [(3, 224, 224), '15.61 M', '2.51 GMac', 16.2]