Google Shows AI Powers in Describing Photos with 94 Percent Accuracy

Services like Google Photos are popular thanks to the integration of AI with the ability to recognize objects in photos, which are then translated into features tagging automatic. This is just one benefit that can be taken from technology image recognition, there are other uses, such as providing oral descriptions for the blind.

How accurately can AI actually recognize objects in images? Based on the recognition of the Google Research team, the accuracy has now reached 93,9 percent. Interestingly, all of this can be enjoyed by all parties developer, not just Google alone.

Yup, Google continues to refine the technology behind machine learning open-source-his, TensorFlow, this time with the "Show and Tell" algorithm that allows developer to train AI in recognizing and identifying various objects in images.

After being trained with three different photos of dogs and their respective descriptions, for example, the system was able to accurately identify and describe photos that it had never seen before. This is also evidence that the new system can understand the object as well as the context more deeply.

Just two years ago, Google's AI could only identify objects with 89,6 percent accuracy. Now, not only has his accuracy increased, but also his ability to develop the information he has acquired – the result of his training with humans – to describe a completely new photo, then express it verbally in a more natural way.

Sumber: Engadget and Google Research.