README.md

metadata

license: cc-by-4.0

FETV

FETV is a benchmark for Fine-grained Evaluation of open-domain Text-to-Video generation

Overview

FETV consist of a diverse set of text prompts, categorized based on three orthogonal aspects: major content, attribute control, and prompt complexity.

Data Statistics

FETV contains 619 text prompts. The data distributions over different categories are as follows

Data Format

All FETV data are all available in the file fetv_data.json. Each line is a data instance, which is formatted as:

{
  "video_id": 1006807024, 
  "prompt": "A mountain stream", 
  "major content": {
       "spatial": ["scenery & natural objects"], 
       "temporal": ["fluid motions"]
     }, 
  "attribute control": {
      "spatial": null, 
      "temporal": null
    }, 
  "prompt complexity": ["simple"], 
  "source": "WebVid", 
  "video_url": "https://ak.picdn.net/shutterstock/videos/1006807024/preview/stock-footage-a-mountain-stream.mp4"
  }

• "video_id" is the video identifier in the original dataset where the prompt comes from.
• "prompt" is the text prompt for text-to-video generation.
• "major content", "attribute control" and "prompt complexity" are the three orthogonal aspects for categorization.
• "source" denotes the original dataset where the prompt comes from, which can be "WebVid", "MSRVTT" or "ours".
• "video_url" is the url link of the reference video.