Learning Representations from Audio-Visual Spatial Alignment

We introduce a novel self-supervised pretext task for learning representations from audio-visual content.