#meta-prompting

ViSMaP introduces an unsupervised approach to summarize hour-long videos effectively using meta-prompting and short-form video datasets, matching supervised model performance without the need for extensive annotations.