Dark Mode
Geometric Shape Classification Targets
Data Science and Analytics
Tags and Keywords
Trusted By
"No reviews yet"
Free
About
A randomly generated image dataset tailored for foundational machine learning practice in both classification and regression. The collection consists of 10,000 square images, each sized at 128 by 128 pixels, featuring a single regular polygon. These shapes include triangles, squares, pentagons, and hexagons. This product is suitable for training models to predict attributes such as the shape type, colour labels, bounding circle coordinates, and rotation angle.
Columns
The accompanying tabular data file (
targets.csv) contains the ground truth attributes for each image:- Index: The record identifier.
- filename: The file name associated with the image located in the
/imagesfolder. - sides: The number of sides defining the polygon (3, 4, 5, or 6).
- bg_color: The label representing the background colour.
- fg_color: The label representing the foreground colour of the polygon.
- bound_circle_x: The X coordinate of the bounding circle surrounding the polygon.
- bound_circle_y: The Y coordinate of the bounding circle surrounding the polygon.
- bound_circle_r: The radius of the bounding circle.
- rotation: The rotation of the polygon, measured in degrees from 0 to 360.
Distribution
The dataset contains a total of 10,000 records. It features regular polygons with 3, 4, 5, or 6 sides, with approximately 2,500 examples for each shape type. There are 10 possible colour labels for both the foreground and background, including "white", "gray", "black", "red", "green", "blue", "yellow", "orange", "purple", and "pink", although the actual colours are slightly varied across images. The data is available in both image (PNG) and tabular (CSV) formats. The maximum expected update frequency for this dataset is never, as it is a static collection.
Usage
This data product is useful for various machine learning challenges. Classification models can be built to predict the polygon shape from the unseen images, or to predict the discrete foreground and background colour labels. Regression models can be trained to predict continuous variables, such as the X and Y coordinates of the bounding circle, its radius, or the polygon's exact rotation in degrees. The full 0-360 degree rotation range provides an interesting test case for regression, particularly because the symmetry of regular polygons means multiple angles can produce an identical image.
Coverage
The dataset is synthetic, focusing solely on geometric generation rules. It does not contain any geographic, temporal, or demographic scope. The bounding circles are constrained to always fall within the limits of the 128x128 pixel images.
License
CC0: Public Domain
Who Can Use It
- Machine Learning Engineers: For benchmarking novel classification and regression architectures using simple visual data.
- Students and Educators: Ideal for introductory lessons on computer vision, feature extraction, and target variable prediction.
- Data Scientists: For quickly prototyping models where geometric features and colour attributes are the primary focus.
Dataset Name Suggestions
- Color Polygon Attributes for ML
- Geometric Shape Classification Targets
- 10,000 Regular Polygon Images
- Simple Vision Dataset
Attributes
Original Data Source: Geometric Shape Classification Targets
QUALITYLoading...