Enabling a mobile app to analyze images is as easy as integrating it with one of the numerous cloud-based object recognition services on the market. But sending files to a remote data center for processing isn’t always efficient because of connectivity constraints and other logistical hurdles.

Now developers have an alternative. Google Inc. on Wednesday open-sourced a family of computer vision algorithms called MobileNets that are specifically designed to run on smartphones and tablets. According to the search giant, the package sidesteps many of the technical obstacles that have historically made it difficult to build object recognition features directly into an app.

The first part of the package involves use of hardware. Computer vision algorithms, like other types of artificial intelligence, typically require a great deal of processing power that can be hard to come by on a mobile device. Google designed each of the 16 models in the MobileNets family with a different performance profile to let developers optimize resource consumption based on their project requirements.

The less fine-grained an app’s object recognition capability needs to be, the more resources can be left for other processes. For added measure, Google has optimized MobileNets to be power efficient in a bid to prevent applications that employ its algorithms from draining the battery too fast.

Once they’ve selected a model, developers can quickly start processing images thanks to the fact that the models in the package come pre-trained. This is a major convenience seeing that getting an artificial intelligence into working order usually requires a great deal of time and effort, not to mention specialized know-how.

Google has trained MobileNets to analyze facial expressions, recognize objects that appear in images and perform several other related tasks. The search giant claims that the package can compete with leading computer vision algorithms despite having been built to run on handsets. It’s designed work with the recently introduced mobile edition of TensorFlow, the company’s popular open-source machine learning engine.

MobileNets should help Google cement its position in the artificial intelligence ecosystem and score more points with mobile developers. But on the flip side, the project could compete with the company’s cloud-based object recognition service, which is a key part of its efforts to make money from the rapid rise of AI.  

Image: Google