发表机构
School of New Media and Communication, Tianjin University; School of Artificial Intelligence, Tianjin University; School of Computer Science and Technology, Tianjin University; School of Sports Training, Tianjin University of Sport; Faculty of Information Engineering and Automation, Kunming University of Science and Technology; School of Electronic Science, National University of Defense Technology; School of Automation, Southeast University; School of Computer Science and Engineering, University of New South Wales（天津大学新媒体与传播学院; 天津大学人工智能学院; 天津大学计算机科学与技术学院; 天津体育学院运动训练学院; 昆明理工大学信息工程与自动化学院; 国防科技大学电子科学学院; 东南大学自动化学院; 新南威尔士大学计算机科学与工程学院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究针对异构互补的目标检测器，提出SDL框架与TGRD方法，通过轨迹引导的定向互蒸馏，实现知识高效整合与跨类别传递，在MS COCO上取得显著性能提升。
AI 中文摘要
目标检测知识分散在独立训练、异构且类别支持互补的检测器中。在社会化学习中，知识存在于一个“社会”中，学习旨在通过集体交流推动该“社会”演化。然而，基于聚合的社会化方法未明确规划知识传递顺序，而渐进式多教师蒸馏虽考虑了顺序，却仍为共享类别空间中的单向学生增强。本文在社会化学习基础上，针对异构、类别专用的目标检测器，提出社会化检测器学习（Socialized Detector Learning, SDL）框架，并设计轨迹引导与互蒸馏（Trajectory-Guided and Reciprocal Distillation, TGRD）方法。TGRD通过保留的特征对齐残差估计定向的检测器间知识传递难度（Inter-Detector Transfer Difficulty, IDTD），预计算固定分数表，贪婪构造载体轨迹；沿该轨迹，知识逐步整合为联合类别载体，再通过互传递返回各专家。条件代理证书分析表明，在给定假设下，渐进式证书不大于聚合目标对应证书。在MS COCO数据集上，采用4个异构专家与2种载体初始化设置，最终载体在两种设置下均比 epoch 匹配的同步聚合对照高出2.6 AP；互蒸馏检测器在先前不支持的类别上达到20.8-28.4 AP，同时保持与原始专家特定性能相差不超过1.3 AP。这些结果支持“感知顺序的渐进整合后接互传递”是检测器“社会”演化的可行机制。
英文摘要
Object detection knowledge is fragmented across independently trained, heterogeneous detectors with complementary category supports. In socialized learning, this knowledge resides in a society, and learning aims to evolve the society collectively through exchange. However, aggregation-based socialization does not explicitly plan transfer order, whereas progressive multi-teacher distillation considers order but remains a one-way student enhancement in a shared category space. Building on Socialized Learning, we formulate Socialized Detector Learning (SDL) for heterogeneous, category-specialized object detectors and propose Trajectory-Guided and Reciprocal Distillation (TGRD).TGRD estimates directed operational Inter-Detector Transfer Difficulty (IDTD) from held-out feature-alignment residuals, precomputes a fixed score table, and greedily constructs a carrier trajectory. Along the trajectory, knowledge is progressively consolidated into a union-category carrier and then returned to experts through reciprocal transfer. A conditional proxy-certificate analysis shows that, under stated assumptions, the progressive certificate is no larger than an aggregated-target counterpart. On MS COCO with four heterogeneous experts and two carrier initializations, final carriers outperform epoch-matched simultaneous aggregation controls by 2.6 AP in both settings. Reciprocal detectors attain 20.8--28.4 AP on previously unsupported categories while remaining within 1.3 AP of original expert-specific performance. These results support order-aware progressive consolidation followed by reciprocal transfer as a viable mechanism for detector-society evolution.
Comments12 pages; supplementary material included
