TA2N: Two-Stage Action Alignment Network for Few-Shot Action Recognition

Few-shot action recognition aims to recognize novel classes (query) using just a few samples (support). The majority of current approaches follow the metric learning paradigm, which learns compare similarity between videos. Recently, it has been observed that directly measuring this is not ideal since different instances may show distinctive temporal distribution, resulting in severe misalignment issues across query and support In paper, we arrest problem from two distinct aspects -- duration evolution misalignment. We address them sequentially through Two-stage Action Alignment Network (TA2N). first stage locates by affine transform, warps each video feature its while dismissing action-irrelevant (e.g. background). Next, second coordinates match spatial-temporal performing temporally rearrange spatially offset prediction. Extensive experiments on benchmark datasets potential proposed method achieving state-of-the-art performance for few-shot recognition.