Skip to content

MiraData

A long-video dataset with structured captions

Visit official site

What MiraData is

The dataset is described as a large-scale collection of long video clips with structured captions.

What you can do with it

  • Train a model for video understanding
  • Test a video generation algorithm
  • Evaluate long video generation

Application scenarios

  • Download metadata and video samples
  • Split clips with the provided scripts
  • Teach video processing and captioning

Pros & cons

✓ What we like

  • The material lists 17 evaluation metrics from six angles
  • Metadata can be downloaded from Google Drive or HuggingFace Dataset

! What to watch out for

  • The described clips average 72 seconds
  • Captions are generated with GPT-4V in the documented flow

FAQ

How long are the clips?

The material states the video clips average 72 seconds.

How are captions produced?

The material describes multi-dimensional captions generated with GPT-4V.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review