Unsupervised Few-Shot Action Recognition via Action-Appearance Aligned Meta-Adaptation
This addresses the problem of reducing annotation costs for video action recognition in computer vision, though it is incremental as it builds on existing meta-learning and unsupervised techniques.
The paper tackles unsupervised few-shot action recognition by introducing MetaUVFS, a method that uses over 550K unlabeled videos and a novel alignment module to train without base-class labels, achieving competitive or superior performance compared to supervised state-of-the-art methods on benchmarks like HMDB51, UCF101, and Kinetics100.
We present MetaUVFS as the first Unsupervised Meta-learning algorithm for Video Few-Shot action recognition. MetaUVFS leverages over 550K unlabeled videos to train a two-stream 2D and 3D CNN architecture via contrastive learning to capture the appearance-specific spatial and action-specific spatio-temporal video features respectively. MetaUVFS comprises a novel Action-Appearance Aligned Meta-adaptation (A3M) module that learns to focus on the action-oriented video features in relation to the appearance features via explicit few-shot episodic meta-learning over unsupervised hard-mined episodes. Our action-appearance alignment and explicit few-shot learner conditions the unsupervised training to mimic the downstream few-shot task, enabling MetaUVFS to significantly outperform all unsupervised methods on few-shot benchmarks. Moreover, unlike previous few-shot action recognition methods that are supervised, MetaUVFS needs neither base-class labels nor a supervised pretrained backbone. Thus, we need to train MetaUVFS just once to perform competitively or sometimes even outperform state-of-the-art supervised methods on popular HMDB51, UCF101, and Kinetics100 few-shot datasets.