The task of dense video captioning is to generate descriptions for multiple events in an untrimmed video. Previous works generate descriptions for each event, and then concatenate the individual sentence descriptions into a final paragraph, often leading to inconsistency and redundancy problems. To overcome these limitations, we propose a context-aware model that leverages context event information from both past and future events to provide conditional dependence on the current event description. Moreover, context event information is encoded by a strong pre-trained context encoder and is incorporated into the captioning module through a gate-attention mechanism. Experimental results on both YouCookII and ActivityNet datasets demonstrate that our model outperforms the existing context-aware and pre-trained model by a margin.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.