基础模型先验在特征空间中增强目标聚焦以用于无源目标检测
Foundation Model Priors Enhance Object Focus in Feature Space for Source-Free Object Detection
用 OV-SAM 生成的类无关二值掩码作为空间先验，直接正则化检测器特征空间而非仅修伪标签，这一「基础模型先验约束特征」的思路可迁移到其他无源域自适应检测、分割任务中抑制背景杂波响应。
原文依据 Using class-agnostic binary masks derived from OV-SAM, SPAR promotes structured and foreground-focused activations
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论当前无源目标检测（SFOD）的最优方法通常依赖平均教师自标注。然而，域偏移往往会削弱检测器维持强目标聚焦表征的能力，导致在背景杂波上出现高置信度激活，并使检测头产生不可靠的伪标签。已有工作主要致力于改进伪标签，却忽视了强化特征空间本身这一根本需求。我们提出 FALCON-SFOD（Foundation-Aligned Learning with Clutter Suppression and Noise Robustness，即融合杂波抑制与噪声鲁棒性的基础模型对齐学习），这是一个旨在增强域偏移下目标聚焦自适应的框架。它由两个互补组件构成。SPAR（Spatial Prior-Aware Regularization，空间先验感知正则化）利用视觉基础模型的泛化能力来正则化检测器的特征空间。SPAR 使用由 OV-SAM 导出的类无关二值掩码，通过引导网络关注目标区域，促进结构化且聚焦前景的激活。IRPL（Imbalance-aware Noise Robust Pseudo-Labeling，不平衡感知的噪声鲁棒伪标签）通过促进严重前景-背景不平衡下的平衡且抗噪学习，对 SPAR 形成补充。在一项将这些设计与更紧的定位与分类误差界相联系的理论分析的指导下，FALCON-SFOD 在 SFOD 基准上取得了有竞争力的性能。我们的代码可在 https://github.com/Sairam13001/FALCON-SFOD 获取。
英文原文摘要 1522 字符
Current state-of-the-art approaches in Source-Free Object Detection (SFOD) typically rely on Mean-Teacher self-labeling. However, domain shift often reduces the detector's ability to maintain strong object-focused representations, causing high-confidence activations over background clutter and unreliable pseudo-labels from the detection head. While prior works mainly refine the pseudo-labels, they overlook the underlying need to strengthen the feature space itself. We propose FALCON-SFOD (Foundation-Aligned Learning with Clutter Suppression and Noise Robustness), a framework designed to enhance object-focused adaptation under domain shift. It consists of two complementary components. SPAR (\underline S patial \underline P rior-\underline A ware \underline R egularization) leverages the generalization strength of vision foundation models to regularize the detector's feature space. Using class-agnostic binary masks derived from OV-SAM, SPAR promotes structured and foreground-focused activations by guiding the network toward object regions. IRPL (\underline I mbalance-aware Noise \underline R obust \underline P seudo-\underline L abeling) complements SPAR by promoting balanced and noise-tolerant learning under severe foreground-background imbalance. Guided by a theoretical analysis that connects these designs to tighter localization and classification error bounds, FALCON-SFOD achieves competitive performance across SFOD benchmarks. Our code is available in https://github.com/Sairam13001/FALCON-SFOD.