Google’s cloud computing tools include a robust machine vision service that can identify the contents of images. If a picture contains a blue shirt, the AI is capable of recognising that and extracting information about the object. In this instance, it might return tags like “blue” and “shirt” but also “clothes” or “buttons.”
Cloud storage provider Box is now linking up with Google to access this technology. The company will incorporate Google’s computer vision-enabled cloud into its own product. Users will be able to search through their files by entering visual details of what they contain.
Google’s AI will do the rest, looking for the photos that best match the entered keywords. No manual tagging of photos will be required as the AI can already “see” what images look like. For customers and businesses, this will make uploading and organising large quantities of images much easier. Less tagging will be needed as relevant phrases can be automatically extracted.
Computer vision technology applied in this form is one of the most obvious ways AI can simplify tedious work for humans. What could have taken hours to achieve manually can now be performed on-demand in the cloud, letting humans get back to work creating more content. The organisation and subsequent retrieval of the images is left to the storage system itself.
Google’s computer vision system is so far advanced because it’s been online for years. It’s acquired a large amount of experience and training that newer rivals can’t yet reproduce. Its technical accomplishments make it an attractive option to companies like Box. The firm can improve its product without having to develop costly machine learning algorithms on its own.
Box recognised that Google’s not the only player in the AI-enhanced cloud though. Business Insider reports the company will collaborate with other search partners in the long-term expansion of its search feature.
Each AI processes data differently and delivers unique results. While Google Compute Engine has proved to be adept at extracting generic tags from photos, rival solutions might be better suited to application-specific scenarios such as recognition of distinct objects like people and documents.
