A 3 A^{3} -FPN: Asymptotic Content-Aware Pyramid Attention Network for Dense Visual Prediction

Meng’en Qin, Yu Song, Quanling Zhao, Xiaodong Yang, Yingtao Che, Xiaohui Yang · 2026

Read A 3 A^{3} -FPN: Asymptotic Content-Aware Pyramid Attention Network for Dense Visual Prediction by Meng’en Qin, Yu Song, Quanling Zhao, Xiaodong Yang, Yingtao Che, Xiaohui Yang (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/arxiv-2604-10210.txt.

Description

Qin and five colleagues propose A3-FPN, a pyramid network for dense tasks that handles object scale shifts. Current pyramid networks miss telling cues and small objects. A3-FPN uses a disentangled design with content-aware focus and a spread column network for global exchange across levels. Fusion gathers nearby content for spot-wise offsets and weights in context-aware sampling. On COCO and Cityscapes, A3-FPN joins leading CNN and transformer models, scoring 49.6 mask AP on COCO and about 86 mean overlap over union there.

Catalog record

CREATORMeng’en Qin, Yu Song, Quanling Zhao, Xiaodong Yang, Yingtao Che, Xiaohui Yang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE68426
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.10210

Browse the full library · About the Tower of Babel