notesum.ai

Published at December 9

A Lightweight U-like Network Utilizing Neural Memory Ordinary Differential Equations for Slimming the Decoder

cs.CV

cs.AI

eess.IV

Released Date: December 9, 2024

Authors: Quansong He¹, Xiaojun Yao, Jun Wu, Zhang Yi, Tao He

Aff.: ¹College of Computer Science, Sichuan University, Chengdu, China

Arxiv: http://arxiv.org/pdf/2412.06262v1

Dataset	Model	Params(M)	GFLOPs	mIoU	DSC \bigstrut
PH2	Swin-UNet Cao et al. (2022)^∗	25.86	5.86	0.872	0.927 \bigstrut
	UNet++ Zhou et al. (2018)^∗	25.66	28.77	0.883	0.936
	SegNetr Cheng et al. (2023)^∗	12.26	10.18	0.905	0.948 \bigstrut
	Att-UNet Oktay et al. (2018)^†	34.88	66.63	0.858	0.903
	Att-UNet (EED)	18.91	17.34	0.905±0.003	0.950±0.002 \bigstrut
	UNeXt Valanarasu and Patel (2022)^∗	1.47	0.57	0.884	0.936
	UNeXt (LMD)	0.88	0.37	0.896±0.003	0.945±0.002 \bigstrut
	UNet Ronneberger et al. (2015)^∗	13.40	31.12	0.878	0.919
	UNet (HD)	9.48	15.11	0.883±0.005	0.938±0.003 \bigstrut
ISIC2017	FAT-Net Wu et al. (2022)^†	88.87	24.63	0.765	0.850 \bigstrut
	MobileViTv2 Mehta and Rastegari (2022)^⋆	1.87	0.70	0.787	0.881
	TransFuse Zhang et al. (2021)^⋆	26.16	11.50	0.792	0.884 \bigstrut
	EGE-UNet Ruan et al. (2023)^⋆	0.053	0.072	0.798	0.888
	EGE-UNet (EED)	0.041	0.077	0.797±0.001	0.887±0.001 \bigstrut
	UNet Ronneberger et al. (2015)^⋆	13.40	31.12	0.770	0.870
	UNet (LMD)	9.45	14.89	0.773±0.001	0.872±0.001 \bigstrut
	MALUNet Ruan et al. (2022)^⋆	0.177	0.085	0.788	0.881
	MALUNet (HD)	0.101	0.095	0.786±0.003	0.880±0.002 \bigstrut
ISIC2018	UTNetV2 Gao et al. (2022)^⋆	12.80	15.50	0.790	0.883 \bigstrut
	SANet Wei et al. (2021)^⋆	23.9	5.96	0.795	0.886
	TransFuse Zhang et al. (2021)^⋆	26.16	11.50	0.806	0.893 \bigstrut
	MALUNet Ruan et al. (2022)^⋆	0.177	0.085	0.803	0.890
	MALUNet (EED)	0.098	0.082	0.806±0.002	0.893±0.001 \bigstrut
	EGE-UNet Ruan et al. (2023)^⋆	0.053	0.072	0.809	0.895
	EGE-UNet (LMD)	0.041	0.077	0.813±0.005	0.897±0.003 \bigstrut
	UNeXt-S Valanarasu and Patel (2022)^⋆	0.32	0.10	0.791	0.883
	UNeXt-S (HD)	0.156	0.079	0.784±0.003	0.879±0.002 \bigstrut