Methods › General › Attention Modules › Spatially Separable Self-Attention

Spatially Separable Self-Attention

2 papers tagged archive 2025-07-28

Introduced by Xiangxiang Chu et al. in Twins: Revisiting the Design of Spatial Attention in Vision Transformers

archive 2025-07-28 Description, source and code snippet are the archive's method entry.

Spatially Separable Self-Attention, or SSSA, is an attention module used in the Twins-SVT architecture that aims to reduce the computational complexity of vision transformers for dense prediction tasks (given high-resolution inputs). SSSA is composed of locally-grouped self-attention (LSA) and global sub-sampled attention (GSA).

Formally, spatially separable self-attention (SSSA) can be written as:

𝐳̂_(i j)ˡ= LSA ( LayerNorm (𝐳_(i j)ˡ⁻¹))+𝐳_(i j)ˡ⁻¹

𝐳_(i j)ˡ=FFN(LayerNorm(𝐳̂_(i j)ˡ))+𝐳̂_(i j)ˡ

𝐳̂ˡ⁺¹= GSA ( LayerNorm (𝐳ˡ))+𝐳ˡ

𝐳ˡ⁺¹= FFN ( LayerNorm (𝐳̂ˡ⁺¹))+𝐳̂ˡ⁺¹

i ∈{1,2, …., m}, j ∈{1,2, …., n}

where LSA means locally-grouped self-attention within a sub-window; GSA is the global sub-sampled attention by interacting with the representative keys (generated by the sub-sampling functions) from each sub-window 𝐳̂_(i j) ∈ℛ^(k₁ ×k₂ ×C) . Both LSA and GSA have multiple heads as in the standard self-attention.

PaperSource

Papers archive 2025-07-28

2 shown of 2, newest first. Repository counts are the archive's code-links table. A Syntology line states what Syntology ran from that paper's harvested code; it is per sample and not a correctness claim.

Tasks archive 2025-07-28

7 tasks the archive attaches to papers tagged with this method, by distinct papers. A task without a page in the catalog is plain text.

TaskPapers
Avg1
Benchmarking1
Fact Checking1
Fact Verification1
Image Classification1
Retrieval1
Semantic Segmentation1

Usage over time archive 2025-07-28

Papers per year tagged with Spatially Separable Self-Attention: 2021 to 2023, peak 1 1 0 2021: 1 paper 2021 2022: 0 papers 2022 2023: 1 paper 2023
Papers per year the archive tags with this method, by the paper's archive date (2 dated). Bars are counts, not a trend claim.

Components: the archive holds no method-to-method composition, so PwC's Components table cannot be rebuilt; the Papers list carries no Results column for the same reason (the archive does not join its leaderboard rows to method tags).

Categories archive 2025-07-28

Attention Modules

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections