Geometric Shapes Dataset

Dataset Description

Dataset Overview

Geometric Shapes Dataset is a synthetic collection containing images of various geometric shapes overlaid with random text. Each image displays a polygon (or only text) on a randomly colored background, with a short random string partially occluding the shape. The dataset is designed for shape classification, image recognition, and robustness testing of computer‑vision models.

Supported Tasks and Leaderboards

• Image Classification: The main task is multi‑class image classification, aiming to identify the shape type in each image.

Data Instances

Each data instance includes:

• An image (50 × 50 px, RGB)
• A label indicating the shape type

Data Fields

• image: a 50 × 50 px RGB image stored as a NumPy array.
• label: a string indicating the shape type. Labels correspond to the following shapes:
  • "1": No shape (only random text on a colored background)
  • "2": Circle‑like shape (approximated by a 100‑sided polygon)
  • "3": Triangle
  • "4": Square
  • "5": Pentagon

Each image contains:

1. A randomly colored background
2. The designated geometric shape (except label "1"), filled with a different random color
3. A short (4‑character) random alphanumeric text overlaid on top, partially occluding the shape

Note: The “circle” (label "2") is approximated by a 100‑sided polygon and appears circular at the given resolution.

Data Splits

The dataset is divided into training (70 %), validation (10 %), and test (20 %) sets.

Dataset Creation

Rationale

The dataset provides a simple, controlled environment for testing image‑classification models, especially when geometric shapes are partially occluded by text.

Source Data

Data are generated synthetically using custom Python scripts; no external data sources are used.

Annotation

Labels are generated automatically during image creation.

Personal and Sensitive Information

The dataset contains no personal or sensitive information.

Known Limitations

• The dataset includes only a predefined set of shapes.
• Image resolution is fixed at 50 × 50 px.
• Text overlay is always present, which may not reflect all real‑world scenarios.

License

The dataset is released under the MIT License.