Datasets
Creating, extending and curating training data for machine learning and AI applications.
Data Solutions
Custom datasets and augmentation for GeoAI and machine learning.
Image Data Augmentation
We extend existing training data with computer vision methods to improve classification and segmentation accuracy, for photographs as well as satellite and aerial imagery. Using PyTorch and TensorFlow, we generate additional image data or build augmentation pipelines tailored to your data. This covers general photography, satellite imagery, multispectral data and time series; combining acquisitions from different sensors and dates makes models more robust across imaging conditions. These are the same techniques we use in our own segmentation projects on Sentinel-2, orthophoto and aerial data.
Synthetic Data Generation
Where real data is scarce, unbalanced or sensitive, we generate synthetic training data with CGI tools such as Blender and GRASS GIS and with generative models (GANs) built in PyTorch or TensorFlow. Synthetic data lets you train for rare cases and controlled conditions, and it avoids exposing production data. We use synthetic datasets in our own training courses for exactly that reason.
Custom Data Collection
We design and build data collection pipelines for machine learning and AI, including web scraping, API integration and sensor data acquisition, with validation and documentation so that datasets remain reproducible. Reference work includes a curated Holocene landscape-factor dataset, built through web scraping and data wrangling for a joint Max Planck and Helmholtz Institute research project.
Ready to transform your geospatial data?
We reply within one business day.