2D Shape Dataset of Proximity-Based Spatial Relation Concepts (GeoConShapes)
Description
This dataset contains synthetic 2D images that depict basic spatial relation concepts such as ‘alone’, ‘close’, ‘far’, and ‘overlap’. The images include simple geometric shapes, artificial cracks and voids, and semantically segmented real-world objects (e.g., cats, birds, plants). Each image is labeled based on the spatial arrangement of the objects it contains. The data was generated using a Python script (available here: https://github.com/randoba/geoconshapes), which allows for the addition of new object types and even further spatial configurations with some coding effort. The dataset is intended for training and evaluating machine learning models to understand basic proximity relationships. The folder structure and naming convention reflect the spatial relation and object types in each image. This dataset is a result of Project C2279767 that has been implemented with the support provided by the Ministry of Culture and Innovation of Hungary from the National Research, Development and Innovation Fund, financed under the KDP-2023 funding scheme.
Files
Steps to reproduce
1. Clone or download the repository from https://github.com/randoba/geoconshapes 2. Follow the instructions in the README to set up the Python environment 3. Run the script generate_geocon_shapes.py to generate the dataset. You can modify parameters in the script to customize the output
Institutions
- Eotvos Lorand Tudomanyegyetem Informatikai Kar