. 2022 Jul 8;3(7):100520. doi: 10.1016/j.patter.2022.100520

Table 2.

Comparison of different spatial information fusion modules

Spatial operation	Block	Spatial	Channel	Resolution sensitive	Params	FLOPs
Spatial projection	MLP-Mixer¹⁵	specific	agnostic	true	$O (H^{2} W^{2})$	$O (H^{2} W^{2} C)$
	ResMLP⁶⁰	specific	agnostic	true	$O (H^{2} W^{2})$	$O (H^{2} W^{2} C)$
	FeedForward⁶¹	specific	agnostic	true	$O (H^{2} W^{2})$	$O (H^{2} W^{2} C)$
	gMLP⁶⁷	specific	agnostic	true	$O (H^{2} W^{2})$	$O (H^{2} W^{2} C)$
	Sparse-MLP(MoE)⁸⁸	specific	agnostic	true	$O (H^{2} W^{2})$	$O (H^{2} W^{2} C)$
	CCS⁹¹	agnostic	group specific	true	$O (H W)$	$O (H W log (H W) C)$
	RepMLPNet⁹²	specific	group specific	true	$O (H^{2} W^{2} + C^{2})$	$O (H^{2} W^{2} C)$
Axial projection	RaftMLP⁷⁴	specific	agnostic	true	$O (H^{2} + W^{2})$	$O (H W C (H + W))$
	ViP⁷¹	specific	specific	true	$O (H^{2} + W^{2} + C^{2})$	$O (H W C (H + W + C))$
	Sparse MLP⁷³	specific	specific	true	$O (H^{2} + W^{2} + C^{2})$	$O (H W C (H + W + C))$
	DynaMixer⁷⁵	specific	specific	true	$O (H^{3} + W^{3} + C^{2})$	$O (H W C (H^{2} + W^{2} + C))$
	WaveMLP⁷⁶	specific	specific	false	$O (2 L^{2} + C^{2})$	$O (H W C (L + L + C))$
	MorphMLP⁷⁷	specific	specific	false	$O (2 L^{2} + C^{2})$	$O (H W C (L + L + C))$
Shifting & channel projection	S2MLP⁷⁸	agnostic	specific	false	$O (C^{2})$	$O (H W C^{2})$
	S2MLPv2⁷⁹	agnostic	specific	false	$O (C^{2})$	$O (H W C^{2})$
	AS-MLP⁸⁰	agnostic	specific	false	$O (C^{2})$	$O (H W C^{2})$
	CycleMLP⁸³	agnostic	specific	false	$O (C^{2})$	$O (H W C^{2})$
	HireMLP⁸⁶	agnostic	specific	false	$O (C^{2})$	$O (H W C^{2})$
	MS-MLP⁸⁷	agnostic	specific	false	$O (C^{2})$	$O (H W C^{2})$
	ActiveMLP⁸⁵	agnostic	specific	false	$O (C^{2})$	$O (H W C^{2})$

H, W, and C are the feature map’s height, width, and channel numbers, respectively. L is the local window size. “Spatial” refers to whether feature extraction is sensitive to the spatial location of objects, “specific” means true, while “agnostic” means false. “Channel” refers whether weights are shared between channels, “agnostic” shares weights between all channels, “group specific” shares weights between groups, and “specific” does not share. “Resolution sensitive” refers to whether the module is resolution sensitive.