Google announces that its visual search engine Google Lens can now identify more than a billion different objects, which is 4,000 times the number of 250,000 objects at the time of Lens’ release in 2017.
Aparna Chennapragada, vice president of Google’s Lens and AR division, says in a blog post that Lens has simply been further trained to recognize objects. The training is done with Google’s open source deep learning system DeepMind. Texts on objects are recognized with an OCR engine and any reading errors are corrected by the model which also corrects typos in Google Search.
According to The Verge, the number of a billion items comes from Google Shopping. This mainly concerns objects that are actually for sale. However, plants, animals and buildings are also categories that Lens can deal with.
Google debuted Lens at its I/O show in 2017. Initially, the service was integrated into Assistant and the Photos app, but a standalone Lens app is now also available to download. Google Lens has recently become available on iOS outside the Photo app; the function is now also within the Google Search app.