Skip to content
DraconiticResearch
Search Sources API
203,203 studies31 sourcesUpdated 3 min ago
Peer-reviewedCC BY 4.0PLOS One

PSAFM: A parameter-free spatial attention fusion module

  • Guangjun Liu1
  • Xiaoping Xu2
  • Bingkun Zhou3
  • Feng Wang4
  • Yike Li
4 affiliations
  1. 1School of Ecological Environment and Chemical Engineering, Xi’an University of Technology, Xi’an, P. R. China
  2. 2School of Mathematics, Xi’an University of Technology, Xi’an, P. R. China
  3. 3School of Mathematics and Statistics, Xi’an Jiaotong University, Xi’an, P. R. China
  4. 4Department of Thyroid and Breast Surgery, Xijing Hospital of Air Force Military Medical University, Xi’an, P. R. China
Open at sourcePDFDOICite

Abstract

Attention mechanisms improve convolutional neural networks (CNNs) by emphasizing informative features, but many existing modules introduce additional parameters, convolutional or fully connected layers, and non-negligible computational overhead. These costs may limit their use in lightweight or plug-and-play CNN architectures. To address this issue, this paper proposes a parameter-free spatial attention fusion module (PSAFM). Specifically, the pointwise average- and max-pooling branches are combined by deterministic weighted fusion, after which channel statistics and a Tanh activation are used to compute adaptive three-dimensional attention weights. This enables the module to strengthen feature responses without adding trainable parameters. We evaluate PSAFM on CIFAR-10 and CIFAR-100 using ResNet, Pre-ResNet, and MobileNetV2 backbones. On CIFAR-10, PSAFM achieves the best accuracy in 8 of 9 tested backbone settings and improves the original backbones by 0.32 to 1.81 percentage points. On CIFAR-100, PSAFM achieves the best accuracy in 6 of 9 settings and improves the original backbones by 0.05 to 1.68 percentage points. In these experiments, PSAFM is compared with representative attention modules. The results show that its parameter count remains essentially unchanged and that only a small FLOPs overhead is introduced. In other words, when model compactness is important, PSAFM is a simple and efficient attention module that improves CNN feature representation.

Related

  1. MSCF-DM: center-site-aware attention and gated multi-scale fusion network for cross-species DNA methylation site prediction
    Xin Wei, Siqin Hu
    BMC GenomicsPeer-reviewedCC BY 4.0Genetics & genomicsCrossref
  2. DHAFGan: A dense hybrid attention fusion generative adversarial network for infrared and visible image fusion
    Qiong Hong, Zhonghua Xu, Dongli Qin +2
    PLOS OnePeer-reviewedCC BY 4.0Full textNeurosciencePLOS
  3. Menstrual cycle and hormonal contraceptive intake alter the electrophysiological signature of selective auditory spatial attention
    M.-C. Schlueter, M. Pinnow, M. Bruene +1
    bioRxivPreprintAll rights reservedFull textNeurosciencebioRxiv
  4. VAD-EmotionNet: a spatial-spectral attention network for EEG emotion recognition in the 3D affective space
    Xiujun Li, Shen Yan, Dan Tong +2
    Cognitive NeurodynamicsPeer-reviewedLicence not statedFull textNeurosciencePMC
  5. Dynamic hypernode injection and graph attention fusion for multimodal sentiment analysis of behavioral signals
    Wenfei Cao, Jian Xu, Yinghao Li +3
    PLOS OnePeer-reviewedCC BY 4.0Full textBioinformatics & computational biologyNeurosciencePLOS

Cite

BibTeXRIS
BibTeX
@article{liu2026psafm,
  title = {{PSAFM: A parameter-free spatial attention fusion module}},
  author = {Guangjun Liu and Xiaoping Xu and Bingkun Zhou and Feng Wang and Yike Li},
  journal = {PLOS One},
  year = {2026},
  month = sep,
  doi = {10.1371/journal.pone.0357958},
  url = {https://doi.org/10.1371/journal.pone.0357958},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - PSAFM: A parameter-free spatial attention fusion module
AU  - Guangjun Liu
AU  - Xiaoping Xu
AU  - Bingkun Zhou
AU  - Feng Wang
AU  - Yike Li
PY  - 2026
DA  - 2026/09/15
JO  - PLOS One
DO  - 10.1371/journal.pone.0357958
UR  - https://doi.org/10.1371/journal.pone.0357958
AB  - Attention mechanisms improve convolutional neural networks (CNNs) by emphasizing informative features, but many existing modules introduce additional parameters, convolutional or fully connected layers, and non-negligible computational overhead. These costs may limit their use in lightweight or plug-and-play CNN architectures. To address this issue, this paper proposes a parameter-free spatial attention fusion module (PSAFM). Specifically, the pointwise average- and max-pooling branches are combined by deterministic weighted fusion, after which channel statistics and a Tanh activation are used to compute adaptive three-dimensional attention weights. This enables the module to strengthen feature responses without adding trainable parameters. We evaluate PSAFM on CIFAR-10 and CIFAR-100 using ResNet, Pre-ResNet, and MobileNetV2 backbones. On CIFAR-10, PSAFM achieves the best accuracy in 8 of 9 tested backbone settings and improves the original backbones by 0.32 to 1.81 percentage points. On CIFAR-100, PSAFM achieves the best accuracy in 6 of 9 settings and improves the original backbones by 0.05 to 1.68 percentage points. In these experiments, PSAFM is compared with representative attention modules. The results show that its parameter count remains essentially unchanged and that only a small FLOPs overhead is introduced. In other words, when model compactness is important, PSAFM is a simple and efficient attention module that improves CNN feature representation.
SN  - 1932-6203
ER  -