Google is opening up its image recognition tech
Yesterday, Google released a limited preview program of their Cloud Vision API. It basically enables developers to build applications that can see, and more importantly understand, the content of images. As their blog post says: “Cloud Vision API quickly classifies images into thousands of categories (e.g., “boat”, “lion”, “Eiffel Tower”), detects faces with associated emotions, and recognizes printed words in many languages. With Cloud Vision API, you can build metadata on your image catalog, moderate offensive content, or enable new marketing scenarios through image sentiment analysis.” And yes, Like every single time such massive technology is opened up you will find a sentence such as “we are very excited to see what happens next”…
Long story short, it seems that all of the goodness that won the imageNet challenge are now available through an API. Instead of illustrating the endless possibilities, Google released a video of a homemade Raspberry Pi based GoPiGo robot calling the Cloud Vision API.
Such milestone release makes me wonder about the evolution of AI and the impact it might have on the AI eco-system.
Firstly, some will blame Google to be one more time a “startup killer” as many were hoping to monetize a similar technology. However, to be honest such realease was predictable. We knew that all the major teams in AI were mastering image recognition, as this Yann LeCun video was showing a live video feed with an accuracy of 10 times per second. Plus, Google has a track record of enabling technology to the public on a frequent basis. In such situation, I personally like to look at the half-full cup instead of the half-empty one. The enabling of embedding such technology in existing website, apps, software, drones or building new project from scratch is just tremendous.
The cloud service format of this release is not a big surprise as well. We know that Google Cloud is a big priority at Google and that they have to catch up with AWS. Therefore in order to do so, they need to push developers to adopt their cloud solution.
However, let’s not forget that many similar services already existed, such as Project Oxford by Microsoft, Wolfram ImageIdentify, CloudSight, Clarifai, MetaMind, Orbeus or the IBM-owned AlchemyAPI, just to name the most famous ones. Will it as good as the existing APIs? I will anxiously be waiting for the pricing, performance and accuracy comparisons, once it will be fully rolled out.
Finally, I am really starting to regret the Google glasses. Matching this API with the Google goggles would have been an amazing source of inspirations for sport, transportation, construction, medicine and plenty of other industries applications.
Opening up the technology behind Google Photos categorization at your disposal is probably the greatest gift Google could offer the Artificial Intelligence community in 2015. Unfortunately, It’s not clear when Cloud Vision API will become generally available but it seems that Christmas came a little early this year for a few lucky ones.