The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider removing the loading script and relying on automated data support (you can use convert_to_parquet from the datasets library). If this is not possible, please open a discussion for direct help.

Dataset Card for BasicShapes1000

Dataset Summary

This is a synthetic dataset containing randomly-generated SVGs with various shapes

Supported Tasks and Leaderboards

NA

Languages

NA

Dataset Structure

The dataset is composed of 4 base domains, plus a 'mixed' domain that is a superset of the other 4:

  • circles - only circles
  • squares - only squares
  • squares_and_circles - circles and squares present in the same svg
  • scer - squares, circles, ellipses, and rectangles present in the same svg
  • mixed - an aggregation of all of the above

Data Instances

There's stuff there

Data Fields

Each example has 4 fields:

  • svg - the raw svg as a string
  • png - a raster rendering of the svg with a white background
  • object_mask - a black/white mask that defines the outlines of the svg objects
  • layer_mask - a greyscale mask that defines layers of svg objects - overlap regions are brighter. Created by making all the objects white and semi-transparent
  • segments - a numpy array in the shape (N,512,512), dtype='bool' where N is the number of svg objects. The array is a mask of each object with True in the area of the object

Data Splits

Train & validation include the layer and object masks, test does not

Dataset Creation

Generated by randomly inserting objects into an SVG.

Curation Rationale

Objects should have at least 50% of their bounding box visible - i.e. no big circle completely obscuring a little circle

Source Data

/dev/urandom

Initial Data Collection and Normalization

NA

Who are the source language producers?

NA

Annotations

see Data Fields

Annotation process

see Data Fields

Who are the annotators?

Imagemagick/pysvg

Personal and Sensitive Information

Unlikely

Considerations for Using the Data

Please do not use for world domination.

Social Impact of Dataset

NA

Discussion of Biases

Dataset is highly biased against triangles and concave shapes

Other Known Limitations

Color selection is pretty limited.

Additional Information

Dataset Curators

Aleks Clark

Licensing Information

CC-BY

Citation Information

Link it I guess?

Contributions

Thanks to @aleksclark for adding this dataset.

Downloads last month
80