Garrulus Dataset Library

Garrulus Dataset Libary (gdl) package provides tools to work with the Garrulus dataset. This particularly includes data pre-processing, benchmarking, task creation such as semantic segementation, object detection, and classification task.

The backend of gdl is based on PyTorch and torchgeo.

Installation

pip install git+https://github.com/garrulus-project/gdl.git

Features and example usage

• Mask label generation from geopackages for field-D (notebook)
• Area of interest (AOI) sampling (sample inside AOI or grid cells). This also allows to sample within predifined train, valid, test grid cells (notebook)
• RasterDataset for semantic segmentation (notebook)
• Semantic segmentation tasks (notebook)

Benchmark datasets

• field-D dataset benchmark

WIP

• Support other fields
• Object detection task
• Classification task
• Multispectral data pre-processing
• More tutorials
• Example usage with PyTorch Lightning
• Large vision models