GRACE: Geometry- and Ray-Aware Camera-Efficient Multi-View Pedestrian Tracking
为解决减少摄像头数量导致的BEV跟踪问题,提出GRACE方法,通过几何和光线感知融合及跟踪恢复技术提高行人跟踪准确性。
为解决减少摄像头数量导致的BEV跟踪问题,提出GRACE方法,通过几何和光线感知融合及跟踪恢复技术提高行人跟踪准确性。
本文提出了一种用于统计数据融合的观测块多预测方法,通过在深度玻尔兹曼机中引入跨块条件,解决传统方法无法同时观察两个结果的问题。
This study addresses the absence of explicit energy models in tabular anomaly detection by revisiting Deep Boltzmann Machines (DBMs). We propose that mean-field energy and reconstruction scores exhibit complementarity, integrating them via a rank fusion strategy to provide a non-redundant perspective for anomaly detection. Experimental results demonstrate significant improvements in AU-ROC across multiple benchmarks, with statistical superiority over established baselines. These findings validate the effectiveness of classical energy-based models on tabular data and underscore their complementary value alongside reconstruction-based approaches. Consequently, this work establishes a novel research direction for tabular anomaly detection by bridging traditional energy modeling with modern evaluation frameworks, offering both theoretical insights and practical performance gains in identifying anomalous instances within structured datasets.
This work addresses the challenges of catastrophic forgetting of old classes and poor learning of new classes in long-tailed class-incremental learning (LT-CIL), which stem from severe data imbalance. To mitigate these issues without introducing additional parameters, the authors propose a task-boundary-aware gradient update modulation mechanism that jointly regulates feature representation updates through adaptive gradient scaling (AGS), confidence-ranked knowledge distillation reweighting (CRK), and a fragility-based entropy gate (FBE). This approach uniquely integrates teacher model uncertainty, prediction confidence ranking, and memory fragility to dynamically adjust both knowledge distillation weights and gradient magnitudes. Evaluated across five LT-CIL settings, the method consistently outperforms the DGR baseline and achieves state-of-the-art task-agnostic accuracy on four mainstream benchmarks, significantly improving performance across new, old, and medium-frequency classes.
This work provides a unified categorical characterization of solution concepts in strategic games, such as Nash equilibria and Pareto-efficient outcomes. By constructing a category of games whose morphisms are mappings between player sets and strategy profiles, and by introducing presheaves valued in the category of sets that assign to each game its solution set, the study formally captures these solution concepts using tools from category theory. It establishes, for the first time, the precise order-theoretic conditions under which such presheaves are well-defined: the Nash equilibrium presheaf is valid when strategy mappings are either order-reflecting or order-embedding and morphisms are relational; the Pareto-efficient presheaf requires strategy mappings to be order-embedding. This framework offers a cohesive categorical semantics for game-theoretic solutions.
为解决减少摄像头数量导致的BEV跟踪问题,提出GRACE方法,通过几何和光线感知融合及跟踪恢复技术提高行人跟踪准确性。
本文提出了一种用于统计数据融合的观测块多预测方法,通过在深度玻尔兹曼机中引入跨块条件,解决传统方法无法同时观察两个结果的问题。
This study addresses the absence of explicit energy models in tabular anomaly detection by revisiting Deep Boltzmann Machines (DBMs). We propose that mean-field energy and reconstruction scores exhibit complementarity, integrating them via a rank fusion strategy to provide a non-redundant perspective for anomaly detection. Experimental results demonstrate significant improvements in AU-ROC across multiple benchmarks, with statistical superiority over established baselines. These findings validate the effectiveness of classical energy-based models on tabular data and underscore their complementary value alongside reconstruction-based approaches. Consequently, this work establishes a novel research direction for tabular anomaly detection by bridging traditional energy modeling with modern evaluation frameworks, offering both theoretical insights and practical performance gains in identifying anomalous instances within structured datasets.
This work addresses the challenges of catastrophic forgetting of old classes and poor learning of new classes in long-tailed class-incremental learning (LT-CIL), which stem from severe data imbalance. To mitigate these issues without introducing additional parameters, the authors propose a task-boundary-aware gradient update modulation mechanism that jointly regulates feature representation updates through adaptive gradient scaling (AGS), confidence-ranked knowledge distillation reweighting (CRK), and a fragility-based entropy gate (FBE). This approach uniquely integrates teacher model uncertainty, prediction confidence ranking, and memory fragility to dynamically adjust both knowledge distillation weights and gradient magnitudes. Evaluated across five LT-CIL settings, the method consistently outperforms the DGR baseline and achieves state-of-the-art task-agnostic accuracy on four mainstream benchmarks, significantly improving performance across new, old, and medium-frequency classes.
This work provides a unified categorical characterization of solution concepts in strategic games, such as Nash equilibria and Pareto-efficient outcomes. By constructing a category of games whose morphisms are mappings between player sets and strategy profiles, and by introducing presheaves valued in the category of sets that assign to each game its solution set, the study formally captures these solution concepts using tools from category theory. It establishes, for the first time, the precise order-theoretic conditions under which such presheaves are well-defined: the Nash equilibrium presheaf is valid when strategy mappings are either order-reflecting or order-embedding and morphisms are relational; the Pareto-efficient presheaf requires strategy mappings to be order-embedding. This framework offers a cohesive categorical semantics for game-theoretic solutions.