{"paper":{"title":"Learning with Semantic Priors: Stabilizing Point-Supervised Infrared Small Target Detection via Hierarchical Knowledge Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A frozen vision foundation model supplies semantic priors to stabilize point-supervised infrared small target detection through hierarchical distillation.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Long Ma, Ping Qian, Weimin Wang, Yuanhang Yao, Zhu Liu","submitted_at":"2026-05-14T04:12:08Z","abstract_excerpt":"Single-frame Infrared Small Target Detection (ISTD) aims to localize weak targets under heavy background clutter, yet dense pixel-wise annotations are expensive. Point supervision with online label evolution reduces annotation cost; however, lightweight CNN detectors often lack sufficient semantics, leading to noisy pseudo-masks and unstable optimization. To address this, we propose a hierarchical VFM-driven knowledge distillation framework that uses a frozen Vision Foundation Model (VFM) during training. We formulate point-supervised learning as a bilevel optimization process: the inner loop "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We propose a hierarchical VFM-driven knowledge distillation framework that uses a frozen Vision Foundation Model (VFM) during training... Experiments on diverse challenging cases across multiple ISTD backbones demonstrate consistent improvements in detection accuracy and training stability.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a frozen general-purpose VFM can reliably supply semantic priors transferable to infrared small targets via SCAM modulation and bilevel optimization without domain-specific adaptation or overfitting to the training distribution.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A hierarchical VFM-driven knowledge distillation method with semantic-conditioned modulation and cluster reweighting stabilizes point-supervised infrared small target detection and improves accuracy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A frozen vision foundation model supplies semantic priors to stabilize point-supervised infrared small target detection through hierarchical distillation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2dfc64c51977f9e2fd4cc20158b156bbb91b576ccde15fc81d3310efdf8b5f09"},"source":{"id":"2605.14346","kind":"arxiv","version":1},"verdict":{"id":"83eec166-015c-4727-bea5-54ce31a9ad45","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:53:14.361858Z","strongest_claim":"We propose a hierarchical VFM-driven knowledge distillation framework that uses a frozen Vision Foundation Model (VFM) during training... Experiments on diverse challenging cases across multiple ISTD backbones demonstrate consistent improvements in detection accuracy and training stability.","one_line_summary":"A hierarchical VFM-driven knowledge distillation method with semantic-conditioned modulation and cluster reweighting stabilizes point-supervised infrared small target detection and improves accuracy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a frozen general-purpose VFM can reliably supply semantic priors transferable to infrared small targets via SCAM modulation and bilevel optimization without domain-specific adaptation or overfitting to the training distribution.","pith_extraction_headline":"A frozen vision foundation model supplies semantic priors to stabilize point-supervised infrared small target detection through hierarchical distillation."},"references":{"count":23,"sample":[{"doi":"","year":2010,"title":"Anal- ysis of new top-hat transformation and the application for infrared dim small target detection.Pattern Recognition, 43(6):2145–2156,","work_id":"a7776eb3-857e-41eb-a18c-a7e19f693675","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"A local contrast method for small infrared target detection.IEEE Transactions on Geoscience and Remote Sensing, 52(1):574–581,","work_id":"a767532b-4763-40fd-8af9-963fbe5bbc2d","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Asymmetric contextual modulation for infrared small target detection","work_id":"ca269583-e519-45a7-97d8-dc1fdfb5c346","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Segment Anything","work_id":"0c3b519e-074e-4112-a638-275301245293","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Dense nested attention network for infrared small target detection.IEEE Transactions on Image Processing, 32:1745–1758,","work_id":"f6ab96a5-2933-40f2-8040-53bf96aae0a9","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":23,"snapshot_sha256":"275e0220fa717406996fc4c1e2e8c76239f5bcfa3a403bb9bb4214237c669892","internal_anchors":2},"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"}