LearningRepresentationsfromAudio-Visual SpatialAlignment

While these approaches learn high-quality representations for downstream tasks such as action recognition, their training objectives disregard spatial cues naturally occurring in audio and visual signals.