{"paper":{"title":"A3-FPN: Asymptotic Content-Aware Pyramid Attention Network for Dense Visual Prediction","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A3-FPN augments feature pyramids with an asymptotically disentangled column network and content-aware attention to capture more discriminative multi-scale features.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Meng'en Qin, Quanling Zhao, Xiaodong Yang, Xiaohui Yang, Yingtao Che, Yu Song","submitted_at":"2026-04-11T13:38:40Z","abstract_excerpt":"Learning multi-scale representations is the common strategy to tackle object scale variation in dense prediction tasks. Although existing feature pyramid networks have greatly advanced visual recognition, inherent design defects inhibit them from capturing discriminative features and recognizing small objects. In this work, we propose Asymptotic Content-Aware Pyramid Attention Network (A3-FPN), to augment multi-scale feature representation via the asymptotically disentangled framework and content-aware attention modules. Specifically, A3-FPN employs a horizontally-spread column network that en"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"A3-FPN can be easily integrated into state-of-the-art CNN and Transformer-based architectures, yielding remarkable performance gains. Notably, when paired with OneFormer and Swin-L backbone, A3-FPN achieves 49.6 mask AP on MS COCO and 85.6 mIoU on Cityscapes.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the asymptotically disentangled column network and content-aware attention modules provide genuine improvements in discriminative feature capture rather than dataset-specific fitting or unmeasured overhead that would not generalize beyond the three evaluated benchmarks.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A3-FPN augments multi-scale representations with asymptotic global interaction and content-aware resampling, delivering gains such as 49.6 mask AP on MS COCO when paired with OneFormer and Swin-L.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A3-FPN augments feature pyramids with an asymptotically disentangled column network and content-aware attention to capture more discriminative multi-scale features.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8566f7ba17d4daea9c2c16af619b101c4ab6766b2c8e12da0bf66e8e27366745"},"source":{"id":"2604.10210","kind":"arxiv","version":1},"verdict":{"id":"4ebfc59c-a737-4d4d-8d11-1de76dc4ab88","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T16:55:41.755210Z","strongest_claim":"A3-FPN can be easily integrated into state-of-the-art CNN and Transformer-based architectures, yielding remarkable performance gains. Notably, when paired with OneFormer and Swin-L backbone, A3-FPN achieves 49.6 mask AP on MS COCO and 85.6 mIoU on Cityscapes.","one_line_summary":"A3-FPN augments multi-scale representations with asymptotic global interaction and content-aware resampling, delivering gains such as 49.6 mask AP on MS COCO when paired with OneFormer and Swin-L.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the asymptotically disentangled column network and content-aware attention modules provide genuine improvements in discriminative feature capture rather than dataset-specific fitting or unmeasured overhead that would not generalize beyond the three evaluated benchmarks.","pith_extraction_headline":"A3-FPN augments feature pyramids with an asymptotically disentangled column network and content-aware attention to capture more discriminative multi-scale features."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.10210/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":66,"sample":[{"doi":"","year":2023,"title":"M. Chen, L. Zhang, R. Feng, X. Xue, J. Feng, Rethinking local and global feature repre- sentation for dense prediction, Pattern Recognition 135 (2023) 109168","work_id":"cbc50938-f756-4ed5-a709-cfa90d453275","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"G. Zhang, Z. Li, C. Tang, J. Li, X. Hu, Cednet: A cascade encoder–decoder network for dense prediction, Pattern Recognition 158 (2025) 111072","work_id":"4f37550e-d9cd-4e4d-b05f-4c8d4cd0215a","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Y . Chen, Z. Zhang, Y . Cao, L. Wang, S. Lin, H. Hu, Reppoints v2: Verification meets regression for object detection, in: Advances in Neural Information Processing Systems, V ol. 33, 2020, pp. 5621–5","work_id":"72c438f7-3d92-4c45-a025-ac764f7ee2af","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"X. Ding, R. Zhang, Q. Liu, Y . Yang, Real-time small object detection using adaptive weighted fusion of efficient positional features, Pattern Recognition 167 (2025) 111717","work_id":"4445e83b-4085-437f-81df-ee50efc24570","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"K. He, G. Gkioxari, P. Dollár, R. Girshick, Mask r-cnn, in: Proc. IEEE Int. Conf. Comput. Vis. (ICCV), 2017, pp. 2961–2969","work_id":"961728bc-a0d5-4cbc-89a4-b83d92becd2b","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":66,"snapshot_sha256":"f9ade67976cbfe308ce6d7841a7a036b3df437cf95116f7513949a329ff080e9","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}