SAFCAR: Structured Attention Fusion for Compositional Action Recognition
This work provides a more effective method for compositional action recognition, which is a significant problem for computer vision systems dealing with complex, multi-component actions.
This paper addresses compositional action recognition, where actions are composed of subjects, atomic-actions, and objects, by proposing a Structured Attention Fusion (SAF) self-attention mechanism. The method effectively combines object detection time-series data with visual context, outperforming current state-of-the-art systems in recognizing novel verb-noun compositions and generalizing efficiently to unseen action categories.
We present a general framework for compositional action recognition -- i.e. action recognition where the labels are composed out of simpler components such as subjects, atomic-actions and objects. The main challenge in compositional action recognition is that there is a combinatorially large set of possible actions that can be composed using basic components. However, compositionality also provides a structure that can be exploited. To do so, we develop and test a novel Structured Attention Fusion (SAF) self-attention mechanism to combine information from object detections, which capture the time-series structure of an action, with visual cues that capture contextual information. We show that our approach recognizes novel verb-noun compositions more effectively than current state of the art systems, and it generalizes to unseen action categories quite efficiently from only a few labeled examples. We validate our approach on the challenging Something-Else tasks from the Something-Something-V2 dataset. We further show that our framework is flexible and can generalize to a new domain by showing competitive results on the Charades-Fewshot dataset.