发表机构
Bowling Green State University; University of Alabama at Birmingham（博林格林州立大学; 阿拉巴马大学伯明翰分校）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
提出差分特征图知识蒸馏（DFM-KD），通过样本间差异传递关系知识，提升基于Transformer的视觉跟踪在资源受限平台的效率与性能。
AI 中文摘要
在自动驾驶感知中，视觉目标跟踪系统必须满足严格的延迟和功耗限制，同时在复杂和动态环境中保持鲁棒性。尽管基于Transformer的跟踪器达到了最先进的精度，但其大量的计算和内存开销阻碍了其在实时、资源受限平台上的部署。为了朝着这一目标迈进，我们提出了差分特征图知识蒸馏（DFM-KD），一种专为基于Transformer的视觉目标跟踪设计的新型关系蒸馏框架。与传统的特征蒸馏方法（如教师和学生特征表示之间的均方误差）最小化逐点差异不同，DFM-KD通过样本间特征差异传递知识，明确对齐特征空间的关系结构。通过蒸馏教师模型如何建模样本间的外观变化和一致性，而不是强制绝对激活的相似性，DFM-KD使学生在批次内更好地捕捉视觉变化的结构动态。因此，蒸馏后的模型展现出增强的特征鲁棒性和改进的跟踪性能。大量实验表明，DFM-KD在跟踪精度和成功率方面始终优于传统的特征级蒸馏方法。
英文摘要
In autonomous driving perception, visual object tracking systems must satisfy stringent latency and power constraints while remaining robust in complex and dynamic environments. Although transformer-based trackers achieve state-of-the-art accuracy, their substantial computational and memory overheads hinder deployment on real-time, resource-constrained platforms. To move toward this goal, we propose Difference Feature Map Knowledge Distillation (DFM-KD), a novel relational distillation framework tailored for transformer-based visual object tracking. Unlike conventional feature distillation methods that minimize point-wise discrepancies (e.g., mean squared error) between teacher and student feature representations, DFM-KD transfers knowledge through inter-sample feature differences, explicitly aligning the relational structure of the feature space. By distilling how the teacher models appearance variation and consistency across samples, rather than enforcing similarity in absolute activations, DFM-KD enables the student to better capture the structural dynamics of visual changes within a batch. As a result, the distilled model exhibits enhanced feature robustness and improved tracking performance. Extensive experiments demonstrate that DFM-KD consistently outperforms conventional feature-level distillation methods in both tracking precision and success rates.
CommentsPublished in the 2026 IEEE Intelligent Vehicles Symposium (IV)
Journal ref2026 IEEE Intelligent Vehicles Symposium (IV), pp. 2032-2037, 2026
DOI：10.1109/IV66570.2026.11624050
