notesum.ai

Published at December 9

EMOv2: Pushing 5M Vision Model Frontier

cs.CV

Released Date: December 9, 2024

Authors: Jiangning Zhang¹, Teng Hu, Haoyang He, Zhucun Xue, Yabiao Wang, Chengjie Wang, Yong Liu, Xiangtai Li, Dacheng Tao

Aff.: ¹Youtu Lab, Tencent, China

Arxiv: http://arxiv.org/pdf/2412.06674v1

Refer to caption

Module	#Params	FLOPs	MPL
MHSA	$4(C+1)C$	$8C^{2}L+4CL^{2}+3L^{2}$	$O(1)$
W-MHSA	$4(C+1)C$	$8C^{2}L+4CLl+3Ll$	$O(Inf)$
Conv	$(Ck^{2}/G+1)C$	$(2Ck^{2}/G)LC$	$O(2W/(k-1))$
DW-Conv	$(k^{2}+1)C$	$(2k^{2})LC$	$O(2W/(k-1))$