The paper introduces the task of dense-captioning events in videos, which involves detecting and describing multiple events occurring simultaneously or consecutively. The authors propose a new model that can identify all events in a video in a single pass while simultaneously generating natural language descriptions for these events. The model includes a variant of an existing proposal module designed to capture both short and long events, and a new captioning module that uses contextual information from past and future events to jointly describe all detected events. The authors also introduce ActivityNet Captions, a large-scale benchmark dataset containing 20k videos with 100k total descriptions, each with unique start and end times. The paper evaluates the model's performance on dense-captioning events, video retrieval, and event localization, demonstrating its effectiveness in detecting and describing events in long or streaming videos.The paper introduces the task of dense-captioning events in videos, which involves detecting and describing multiple events occurring simultaneously or consecutively. The authors propose a new model that can identify all events in a video in a single pass while simultaneously generating natural language descriptions for these events. The model includes a variant of an existing proposal module designed to capture both short and long events, and a new captioning module that uses contextual information from past and future events to jointly describe all detected events. The authors also introduce ActivityNet Captions, a large-scale benchmark dataset containing 20k videos with 100k total descriptions, each with unique start and end times. The paper evaluates the model's performance on dense-captioning events, video retrieval, and event localization, demonstrating its effectiveness in detecting and describing events in long or streaming videos.