New York’s History Comes Alive with AI! ‘OldNYC’ Summons 10,000 Vintage Photos on the Map with GPT-4o!
📰 News Summary
- The ‘OldNYC’ platform, which allows users to explore historical photos of New York, has received a significant AI upgrade. An additional 10,000 photos have been added to the map, bringing the total to 49,000 available for viewing.
- By leveraging OpenAI’s GPT-4o, the platform has introduced a geocoding feature that extracts coordinates from the text descriptions of photos, dramatically enhancing location accuracy.
- Transitioning from Google Maps to OpenStreetMap/MapLibre has not only reduced costs but also allowed for better control over map styling to reflect the aesthetics of the 1930s.
💡 Key Highlights
- Advanced Location Identification with GPT: GPT-4o interprets “names of schools that no longer exist” and “vague descriptions of intersections,” enabling the accurate placement of about 6,000 photos in locations that were previously challenging to pinpoint using conventional methods.
- Dramatic OCR Improvements: Old typewriter text that was previously “garbled” can now be deciphered through a new OCR system powered by GPT-4o-mini. The number of digitized images has surged from 25,000 to 32,000.
- Ditching Google Maps: In response to changes in Google’s pricing model, the shift to OpenStreetMap has made customizations more flexible, allowing for features like removing highways that didn’t exist in the 1930s from the map.
🦈 Curator’s Perspective
The implementation that allows GPT-4o to tackle the painstaking task of pinpointing locations from vintage photo descriptions, achieving an accuracy of 87%, is incredibly cool! The way it reads context to match outdated intersections and school names with OpenStreetMap is a thrilling example of AI breathing new life into legacy data. Seeing GPT-4o-mini effortlessly transform previously “meaningless strings” of text into perfect digital text is a stark reminder of the tech gap between 2015 and 2024!
🚀 What’s Next?
In the future, we can anticipate the integration of AI features that analyze the images themselves, automatically identifying people, buildings, and indoor/outdoor scenes. There’s also potential for photos from other archival collections to be integrated into this AI pipeline!
💬 Haru Same’s Takeaway
The past and future are now connected through AI! It’s like taking a time machine back to New York of yesteryear! 🦈🔥
📚 Terminology
-
Geocoding: The technology that converts textual information about addresses or place names into numerical latitude and longitude data on maps.
-
OCR (Optical Character Recognition): The technology that reads characters from printed documents or images and converts them into digital text data.
-
OpenStreetMap (OSM): An open-data mapping project that anyone can freely participate in, edit, and use.
