Google wants your help training their image recognition algorithms

By | 6th April 2018

Google is all about algorithms and machine learning these days. Android users see this technology first-hand with features like Google Assistant and smart replies. They are constantly working to improve the image recognition algorithms. One shortcoming of the company’s AI systems is the lack of data for emerging markets like India. Google is hoping they can get more human input to fix this.

Currently, the company’s AI systems work pretty well in parts of the world like the U.S. Anurag Batra, product manager at Google, says “We have a very sparse training data set from parts of the world that are not the United States and Western Europe.” Batra is in charge of a project that aims to improve this situation. His team has built an app that asks people to perform easy tasks while simultaneously checking the accuracy of the company’s image recognition and translation algorithms.

google image-recognition

The company is promoting the Crowdsource app in India and other parts of Asia, but anyone can download it. Batra says “Things like what does a sewing machine look like in your world or what does a pair of slippers look like in your world can really help us.” Users verify labels and help transcribe handwriting. Some of the tasks show where the company’s image recognition systems need more work. Images of nuns were labeled as “brides,” for example.

Unlike the Google Opinion Rewards app, there are no monetary rewards for user input. People will receive points, badges, and certificates. Users who collect enough will be invited to join groups with other top contributors. If you’re interested in joining Google’s crowdsourcing efforts, download the app below.

Crowdsource (Free, Google Play) →


Source: WIRED