AI helps add 10k more photos to OldNYC
a day ago
- #historical photography
- #AI integration
- #open source mapping
- The author quietly rebuilt parts of the OldNYC photo viewer, adding 10,000 historic photos, improving location accuracy, and reducing costs.
- Key improvements include better geolocation using AI (GPT-4o) to extract locations from descriptions, and switching to OpenStreetMap for historical accuracy.
- OCR was enhanced with gpt-4o-mini, achieving near-perfect transcription and reducing errors compared to the old custom pipeline.
- The site migrated from Google Maps to OpenStreetMap and MapLibre to avoid costs, allowing removal of modern anachronisms like highways.
- Future plans involve AI extraction of more details, integrating other photo collections, contributing to OpenHistoricalMap, and expanding to other cities.