Methods › General › Attention Mechanisms › BAM

Bottleneck Attention Module

BAM

33 papers tagged archive 2025-07-28

Introduced by Jongchan Park et al. in BAM: Bottleneck Attention Module

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

Park et al. proposed the bottleneck attention module (BAM), aiming to efficiently improve the representational capability of networks. It uses dilated convolution to enlarge the receptive field of the spatial attention sub-module, and build a bottleneck structure as suggested by ResNet to save computational cost.

For a given input feature map X, BAM infers the channel attention s_c ∈ℝ^C and spatial attention sₛ∈ℝ^(H×W) in two parallel streams, then sums the two attention maps after resizing both branch outputs to ℝ^(C×H ×W). The channel attention branch, like an SE block, applies global average pooling to the feature map to aggregate global information, and then uses an MLP with channel dimensionality reduction. In order to utilize contextual information effectively, the spatial attention branch combines a bottleneck structure and dilated convolutions. Overall, BAM can be written as s_c = BN(W₂(W₁GAP(X)+b₁)+b₂)

sₛ = BN(Conv₂^(1 ×1)(DC₂^(3×3)(DC₁^(3 ×3)(Conv₁^(1 ×1)(X))))) s = σ(Expand(sₛ)+Expand(s_c)) Y = s X+X where Wᵢ, bᵢ denote weights and biases of fully connected layers respectively, Conv₁^(1×1) and Conv₂^(1×1) are convolution layers used for channel reduction. DCᵢ^(3×3) denotes a dilated convolution with 3×3 kernel, applied to utilize contextual information effectively. Expand expands the attention maps sₛ and s_c to ℝ^(C×H×W).

BAM can emphasize or suppress features in both spatial and channel dimensions, as well as improving the representational power. Dimensional reduction applied to both channel and spatial attention branches enables it to be integrated with any convolutional neural network with little extra computational cost. However, although dilated convolutions enlarge the receptive field effectively, it still fails to capture long-range contextual information as well as encoding cross-domain relationships.

PaperSource

Papers archive 2025-07-28

30 shown of 33, 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

20 shown of 55 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
Decision Making2
Decoder2
Object2
Object Detection2
Specificity2
Variational Inference2
object-detection2
Action Recognition1
Action Understanding1
Automatic Speech Recognition1
Automatic Speech Recognition (ASR)1
Benchmarking1
Change Detection1
Classification1
Clustering1
Crack Segmentation1
Deep Learning1
Feature Importance1
Fine-grained Action Recognition1
GPU1

Usage over time archive 2025-07-28

Papers per year tagged with BAM: 2018 to 2025, peak 8 8 0 2018: 1 paper 2018 2019: 4 papers 2019 2020: 2 papers 2020 2021: 4 papers 2021 2022: 5 papers 2022 2023: 5 papers 2023 2024: 8 papers 2024 2025: 4 papers 2025
Papers per year the archive tags with this method, by the paper's archive date (33 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 Mechanisms

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