🤖 AI Summary
To address the significant performance degradation of temporal action localization (TAL) under unsupervised domain adaptation (UDA), this paper introduces the first semantic adversarial UDA framework tailored for sparse action detection. Methodologically, it proposes: (1) a local class-sensitive adversarial loss that overcomes the limitations of conventional global distribution alignment—namely, its neglect of action semantics and temporal sparsity; (2) a novel multi-dimensional domain-shift benchmark, EpicKitchens100→CharadesEgo, specifically designed to evaluate sparse TAL under UDA—a previously unaddressed gap; and (3) an integrated architecture jointly modeling sparse action features and enforcing semantically discriminative domain alignment. Experiments across standard UDA settings demonstrate that our approach achieves up to a 6.14% absolute improvement in mean average precision (mAP), substantially outperforming both fully supervised state-of-the-art methods and existing UDA approaches, thereby enhancing robustness of cross-domain action localization.
📝 Abstract
Temporal Action Localization (TAL) is a complex task that poses relevant challenges, particularly when attempting to generalize on new -- unseen -- domains in real-world applications. These scenarios, despite realistic, are often neglected in the literature, exposing these solutions to important performance degradation. In this work, we tackle this issue by introducing, for the first time, an approach for Unsupervised Domain Adaptation (UDA) in sparse TAL, which we refer to as Semantic Adversarial unsupervised Domain Adaptation (SADA). Our contributions are threefold: (1) we pioneer the development of a domain adaptation model that operates on realistic sparse action detection benchmarks; (2) we tackle the limitations of global-distribution alignment techniques by introducing a novel adversarial loss that is sensitive to local class distributions, ensuring finer-grained adaptation; and (3) we present a novel set of benchmarks based on EpicKitchens100 and CharadesEgo, that evaluate multiple domain shifts in a comprehensive manner. Our experiments indicate that SADA improves the adaptation across domains when compared to fully supervised state-of-the-art and alternative UDA methods, attaining a performance boost of up to 6.14% mAP.
