DREAM-1K is a video description benchmark dataset released by ByteDance. The related paper results are:Tarsier: Recipes for Training and Evaluating Large Video Description Models". The dataset contains 1,000 annotated video clips (about 10 seconds in length) of varying complexity from 5 different categories, each containing at least one dynamic event that cannot be accurately identified from a single frame. Each video is provided with fine-grained manual annotations covering all events, actions, and motions.

Data source category:

• Live-action movies
• Animated Films
• Stock Video
• YouTube long videos
• TikTok-style short videos