PSAFM: A parameter-free spatial attention fusion module
Guangjun Liu, Xiaoping Xu, Bingkun Zhou, Feng Wang, Yike Li · 2026
Read PSAFM: A parameter-free spatial attention fusion module by Guangjun Liu, Xiaoping Xu, Bingkun Zhou, Feng Wang, Yike Li (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13577465.txt.
Description
Guangjun Liu and four co-authors design an attention module for image networks that adds no new learnable parameters at all. Attention usually costs weights and computation, which hurts lightweight models. Their module fuses average-pooling and max-pooling branches with fixed weights, then derives three-dimensional attention from channel statistics passed through a tanh function.
Tested on CIFAR-10 and CIFAR-100 across ResNet, Pre-ResNet, and MobileNetV2 backbones, it took best accuracy in most settings. Gains over the plain backbones ran from a third of a point to nearly two points. Parameter count stayed essentially unchanged and computing cost rose only slightly. The authors pitch it as a plug-and-play upgrade for efficient networks.
Catalog record
| CREATOR | Guangjun Liu, Xiaoping Xu, Bingkun Zhou, Feng Wang, Yike Li |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 30 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |