MTG-AD
A video dataset and evaluation benchmark released by a Google team for the inline audio description task, annotated jointly by blind and sighted experts, with emphasis on timeline alignment and real-world usage scenarios.
Timeline (last 90 days)
The Google team released the MTG-AD dataset and the MTG-AD-Eval evaluation framework for the inline audio description (inline AD) task.
The MTG-AD dataset contains 36 hours of video, averaging about 5 minutes per video, with a total of 8k+ descriptions, each aligned to the timeline.
The MTG-AD dataset was annotated by blind and sighted audio description experts working in pairs.
The MTG-AD dataset is open-sourced on Hugging Face at google/mind-the-gap-inline-ad.
All Facts (4)
The MTG-AD dataset contains 36 hours of video, averaging about 5 minutes per video, with a total of 8k+ descriptions, each aligned to the timeline.
50%UnverifiedThe MTG-AD dataset was annotated by blind and sighted audio description experts working in pairs.
50%UnverifiedThe Google team released the MTG-AD dataset and the MTG-AD-Eval evaluation framework for the inline audio description (inline AD) task.
50%UnverifiedThe MTG-AD dataset is open-sourced on Hugging Face at google/mind-the-gap-inline-ad.
50%