Show HN: Build ML training datasets from large-scale satellite/aerial imagery (github.com)

A new open-source tool called "label-tiles" has been introduced for creating machine learning (ML) training datasets from large-scale satellite and aerial imagery. This tool allows users to draw labeled bounding boxes on map images sourced from any tile server, enabling the extraction of geospatial data in formats compatible with popular machine learning frameworks like PyTorch and Ultralytics. Users can configure multiple tile servers, utilize hotkey-driven label assignments, and export data in COCO annotation format, GeoJSON, or GeoParquet, making it highly versatile for various ML tasks. The significance of this tool lies in its ability to simplify the labeling process for geospatial datasets, which is often complex and resource-intensive. By facilitating easy access to tiles and enabling collaborative labeling, "label-tiles" aims to streamline the creation of ML-ready datasets without the heavy lifting typically associated with data transfer and management. The tool uses React, FastAPI, and MapLibre for its development and offers features such as drag-and-drop support for GeoTIFF files, making it accessible for both individual users and teams needing to prepare geospatial data for analysis or model training.