Our take on AI image recognition has begun to deliver. During the last 6 months, we built our second neural network (the first one was image upscale) and ran about half of our 25 million public domain images
through this new image recognition engine.
To make it work we had to learn how to recognize objects, art styles, media types, characters, and, to some extent even historical dates. The results AI delivers still require tight human oversight as neural network classification accuracy is still far from perfect. We manually review enormous, strange, and often surreal lists to deliver satisfactory topics and collections. Here is one that was intentionally left as is: Not Quite Shirley Temple.
Our task of manual review makes sense as we don't have to deal with low-quality and obsolete objects reshuffling them as Google has to. Our images survived the test of time and human curation. We add more public-domain media and expand the number of classes to run our AI recognition against. We also replaced the annoying captcha with a meaningful tag-voting on download. Users can vote for relevant AI-recognised tags, so the more time and effort invested, the better results become.
This is only the beginning, or rather the very first version built. And it is really important that it was independently built with donations and subscriptions of users like you. Now, we can compile huge collections and scale up our existing topics to improve search results for you. Here are some AI-driven collections:
13,000+ samples of Northern Renaissance Art
Almost 4,000 Quattrocento Artworks
51,000 Train Stations
Almost 6,000 Cars from1940s
Few Thousands Pre-1920s Trucks
5,000 People wearing Spanish collars (We can add more but why?)
5,000 Fishing Boats
27,000 Barel Artillery images
1,500 Abraham Lincoln Portraits
38,000 Steam Locomotives
And as a cherry on top, 120,000 Military Parades and Ceremonies
Note that we now can distinguish between sailing boats, sailing ships, sailing yachts, warships, steamships, cargo ships, battleships, and so on...
We continue our mission of making every valuable public domain image annotated, tagged, and available for search and download. Right now we run our AI against another 12M images. Another million or so will be added soon, and we repeat the process. Please visit our collections, donate and upgrade if you are not yet done so. We rely on our users to accelerate this process.
Find all recent updates here: GetArchive Progress Report