Multiview Pseudo-Labeling for Semi-supervised Learning from Video
This work addresses the challenge of semi-supervised learning in video for computer vision applications, offering a method that enhances representation learning without extra inference cost.
The paper tackles the problem of learning video representations with limited labeled data by using multiview pseudo-labeling that leverages appearance and motion information to generate reliable pseudo-labels, resulting in substantial performance improvements over supervised baselines on multiple video recognition datasets.
We present a multiview pseudo-labeling approach to video learning, a novel framework that uses complementary views in the form of appearance and motion information for semi-supervised learning in video. The complementary views help obtain more reliable pseudo-labels on unlabeled video, to learn stronger video representations than from purely supervised data. Though our method capitalizes on multiple views, it nonetheless trains a model that is shared across appearance and motion input and thus, by design, incurs no additional computation overhead at inference time. On multiple video recognition datasets, our method substantially outperforms its supervised counterpart, and compares favorably to previous work on standard benchmarks in self-supervised video representation learning.