Dataset Card for Animations Dataset

Dataset Summary

This dataset contains 50,849 animations with their associated metadata, frames, and source images. Each animation consists of multiple frames composed of simple sketch-level drawings, text elements, and potentially embedded images. The dataset provides complete information about each animation, including frame components, source images, timing between frames, and canvas settings. This makes it suitable for various tasks such as animation analysis, generation, and modification. All animations are stored in a structured format that preserves the relationship between component images and their composition within animation frames.

Languages

The dataset is primarily monolingual:

• English (en): Any text elements within animations are predominantly in English.

Dataset Structure

Data Files

The dataset consists of two main types of files:

1. Metadata files:

   • .jsonl.zst compressed files (e.g., animations_part_1.jsonl.zst)
   • Each contains metadata for approximately 1,000 animations
   • Located in the dataset root directory

2. Frame and source image files:

   • .tar.zst compressed tar archives (e.g., files/animations_part_1.tar.zst)
   • Each contains frames and source images for approximately 1,000 animations
   • Located in the files directory

Data Fields

Each animation entry in the metadata files contains the following fields:

• id: Unique identifier (UUID) for the animation.
• settings: Object containing canvas configuration:
  • canvas_width, canvas_height: Dimensions of the animation canvas.
  • fillcolor: Background color of the canvas (if specified).
  • default_font: Default font used for text elements.
  • default_font_size: Default font size.
• images: Array of image resources used in the animation.
• dtimes: Array of time delays between frames in milliseconds.
• frames_data: Array of arrays representing each frame's elements:
  • type_for_loader: Element type (e.g., "text").
  • data: Object containing element properties:
    • type: Element type.
    • centerx, centery: Position coordinates of the element on the canvas.
    • text: Text content (for text elements).
    • font: Font used for the text.
    • size: Font size.
    • rotate_angle: Rotation angle.
    • strokeColor: Stroke color (if specified).
    • fillColor: Fill color (if specified).
    • textColor: Text color (for text elements).
    • align: Text alignment.
    • angle: Angle for the element.
    • radius: Radius (for circular elements).
  • children_data: Array of child elements (if any).

Directory Structure

The frame and source image files are organized within the tar archives as follows:

[uuid]/
├── frames/
│   └── frame_0000.png
│   └── frame_0001.png (if multiple frames)
│   └── ...
└── source_images/
    └── src_image_0000.png (if source images are used)
    └── src_image_0001.png (if multiple source images)
    └── ...

Where [uuid] is the unique identifier for each animation. Note that the source_images directory will be created even if it contains no files, as some animations may not use any source images.

Data Splits

The dataset contains a single split:

Split	Description	Number of Examples
train	Animation data with frames and source images	50,849

Total dataset size: 50,849 entries