HN
Today

AI helps add 10k more photos to OldNYC

Dan Vanderkam has significantly upgraded OldNYC, his historical photo viewer, by integrating modern AI and OpenStreetMap technologies. This overhaul has added 10,000 new images, drastically improved geolocation accuracy, and enhanced optical character recognition for photo descriptions. The project exemplifies how contemporary AI tools and open-source mapping solutions can breathe new life into digital archives, offering a cost-effective and powerful way to make historical data more accessible and engaging.

17
Score
2
Comments
#8
Highest Rank
3h
on Front Page
First Seen
Apr 7, 5:00 PM
Last Seen
Apr 7, 7:00 PM
Rank Over Time
8910

The Lowdown

The OldNYC photo viewer, which maps historical photos of New York City, has undergone a substantial two-year reconstruction, culminating in the addition of 10,000 new photos and significant improvements to its core functionalities. This revitalization leverages advancements in artificial intelligence and the OpenStreetMap ecosystem, making the site more accurate, robust, and cost-efficient.

  • Enhanced Geolocation with GPT and OpenStreetMap: The project now utilizes OpenAI's gpt-4o to interpret complex textual descriptions, successfully geocoding approximately 6,000 additional photos where traditional methods failed. Additionally, the proprietary Google Maps Geocoding API was replaced with OpenStreetMap and historical street datasets, accurately locating images tied to obsolete street intersections.
  • AI-Powered OCR Revolution: The optical character recognition (OCR) system, originally a challenging technical hurdle, was rebuilt using gpt-4o-mini. This upgrade expanded text coverage from 25,000 to 32,000 images and delivered significantly higher accuracy, even with challenging fonts, compared to the previous custom Ocropus pipeline.
  • Migration to OpenStreetMap: Driven by Google Maps' changing pricing model, OldNYC migrated to OpenStreetMap vector tiles and MapLibre. This shift ensures the project's sustainability while also offering benefits like faster rendering, smoother zooming, and complete control over map styling, allowing for the removal of anachronistic modern features.
  • Future Prospects: Looking ahead, potential improvements include using AI to extract more nuanced information from images (e.g., identifying people or buildings), integrating photos from other collections, contributing to OpenHistoricalMap for enhanced historical street data, and developing tools to simplify the creation of similar historical mapping projects for other cities.

This comprehensive update showcases a compelling application of current technology to preserve and present urban history, making historical data more accessible and engaging for enthusiasts and researchers alike.