Step by Step: A Gradual Approach for Dense Video Captioning

Dense video captioning aims to localize and describe events for storytelling in untrimmed videos. It is a conceptually very challenging task that requires concise, relevant, coherent based on high-quality event localization. Unlike simple temporal action localization tasks without overlapping events, dense detecting multiple/overlapping regions order branch out the story. Most existing methods generate numerous candidate proposals then eliminate duplicate ones using proposal selection algorithm (e.g., non-maximum suppression) or directly through box prediction binary classification mechanisms, similar object detection tasks. Despite these efforts, aforementioned approaches tend fail into different stories, hindering captioning. In this paper, we propose SBS, framework with gradual approach addresses challenge of localizing eventually constructs SBS accurately estimates number explicit each snippet detects boundaries context/activities, which are details generating proposals. Based both boundaries, generates encodes context sequence finally sentences describing Our fairly effective thus experimental results show state-of-the-art performance compared methods.