Capturing Rich Behavior Representations: A Dynamic Action Semantic-Aware Graph Transformer for Video Captioning

TL;DR

The paper addresses the不足 of shallow object-behavior representations in video captioning by introducing a dynamic action semantic-aware graph transformer that jointly models long- and short-term latent actions and their semantic cues. It couples a multi-scale temporal modeling module with a visual-action semantic aware module to produce rich action representations, which are integrated through a temporal objects-action graph and refined by a graph transformer. Knowledge distillation keeps inference efficient by transferring learned behavior knowledge to a simpler network. Empirical results on MSVD and MSR-VTT demonstrate significant improvements in standard metrics, highlighting more accurate and descriptive captions that better reflect object dynamics and interactions.

Abstract

Existing video captioning methods merely provide shallow or simplistic representations of object behaviors, resulting in superficial and ambiguous descriptions. However, object behavior is dynamic and complex. To comprehensively capture the essence of object behavior, we propose a dynamic action semantic-aware graph transformer. Firstly, a multi-scale temporal modeling module is designed to flexibly learn long and short-term latent action features. It not only acquires latent action features across time scales, but also considers local latent action details, enhancing the coherence and sensitiveness of latent action representations. Secondly, a visual-action semantic aware module is proposed to adaptively capture semantic representations related to object behavior, enhancing the richness and accurateness of action representations. By harnessing the collaborative efforts of these two modules,we can acquire rich behavior representations to generate human-like natural descriptions. Finally, this rich behavior representations and object representations are used to construct a temporal objects-action graph, which is fed into the graph transformer to model the complex temporal dependencies between objects and actions. To avoid adding complexity in the inference phase, the behavioral knowledge of the objects will be distilled into a simple network through knowledge distillation. The experimental results on MSVD and MSR-VTT datasets demonstrate that the proposed method achieves significant performance improvements across multiple metrics.

Figures (3)

• Figure 1: Qualitative examples of captioning that do not adequately capture the essence of the object's behavior.
• Figure 2: The pipleline of the proposed method. For training, the proposed graph transformer network and the visual-text network are trained simultaneously. In the model inference phase, only the visual-text network is used because the visual-text network has already learned the object behavior knowledge of the whole network through the knowledge distillation process. This approach avoids complex computations and increases the speed of inference.
• Figure 3: Visualization examples for qualitative comparisons between our method and the baseline model (better viewed in color).