面向鲁棒目标检测的分布对齐多模态融合
Distribution-Aligned Multimodal Fusion for Robust Object Detection
冻结预训练检测器、只训练轻量融合模块，并把融合特征对齐到检测器最优工作的预训练分布，这一思路可直接迁移到任何多模态或跨域检测的适配场景，避免针对训练退化过拟合。
原文依据 aligning fused features to the pretrained distribution where the frozen detector performs optimally
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论跨退化泛化仍是 RGB-红外多模态目标检测的关键挑战，尤其当训练数据仅覆盖有限的退化类型时。本文提出一种分布对齐框架，其核心洞见是：将融合特征对齐到预训练分布——即冻结检测器表现最优的分布——而非去适应训练中特定的退化。通过冻结预训练检测器并仅训练一个轻量融合模块，本方法利用互补的红外信息来减小分布偏移，同时保持计算效率。该方法在三个基准上取得最优结果，训练速度提升 4 倍。关键的是，我们证明在泛化到未见场景时，对齐预训练分布显著优于对齐训练退化。
英文原文摘要 878 字符
Cross-degradation generalization remains a critical challenge for RGB-infrared multimodal object detection, especially when training data covers limited degradation types. This paper presents a distribution alignment framework with a key insight: aligning fused features to the pretrained distribution where the frozen detector performs optimally, rather than adapting to training-specific degradations. By freezing the pretrained detector and training only a lightweight fusion module, our approach leverages complementary infrared information to reduce distribution shift while maintaining computational efficiency. The method achieves state-of-the-art results on three benchmarks with 4x faster training. Critically, we demonstrate that aligning to the pretrained distribution substantially outperforms aligning to training degradations when generalizing to unseen scenarios.