/// Project · VISION / NUMISMATICS
Kozeni
Catalogues a coin collection from a photograph of the tray: each coin is found, identified and filed.
notes
Cataloguing coins one by one is slow. Kozeni takes a photo of a whole tray, separates the coins with classic computer vision, and asks a vision model for year, country, denomination and material.
The heavy work runs in a separate Python worker behind a queue, with progress and retries, so a large scan never blocks the interface. Coins can be matched against the Numista catalogue.