C2C: Component-to-Composition Learning for Zero-Shot Compositional Action Recognition
This addresses the challenge of compositional generalization in action recognition for AI systems, enabling recognition of unseen action combinations, though it is incremental as it builds on existing datasets and methods.
The paper tackles the problem of zero-shot compositional action recognition, where models must recognize unseen actions composed of previously observed verbs and objects, by proposing a new benchmark and a Component-to-Composition (C2C) learning method, achieving state-of-the-art results.
Compositional actions consist of dynamic (verbs) and static (objects) concepts. Humans can easily recognize unseen compositions using the learned concepts. For machines, solving such a problem requires a model to recognize unseen actions composed of previously observed verbs and objects, thus requiring so-called compositional generalization ability. To facilitate this research, we propose a novel Zero-Shot Compositional Action Recognition (ZS-CAR) task. For evaluating the task, we construct a new benchmark, Something-composition (Sth-com), based on the widely used Something-Something V2 dataset. We also propose a novel Component-to-Composition (C2C) learning method to solve the new ZS-CAR task. C2C includes an independent component learning module and a composition inference module. Last, we devise an enhanced training strategy to address the challenges of component variations between seen and unseen compositions and to handle the subtle balance between learning seen and unseen actions. The experimental results demonstrate that the proposed framework significantly surpasses the existing compositional generalization methods and sets a new state-of-the-art. The new Sth-com benchmark and code are available at https://github.com/RongchangLi/ZSCAR_C2C.