eBay has announced that it is working on options for users to search for products by image, or by sharing a link containing a product or product picture to the eBay app, both driven by machine learning. The two new features take advantage of computer vision and deep learning in order to train an AI to match images on a page or user-supplied imagery with pictures of actual products on eBay as closely as possible. Naturally, this means that the system’s matching capabilities only have the training that they were built with to go off of for now, but will improve over time as more and more people use the features. The Image Search feature was dreamed up during one of eBay’s company Hack Week competitions.
The Image Search feature allows users to take a picture of a product that they want to find on eBay, or select a picture from their camera roll, then crop that picture to show the product they want. Users can access the feature via the search bar in the eBay app. Once a picture’s results are pulled up, users can scan through visually similar listings, including color variants with some items, to find exactly what they’re looking for. Listings are ordered by how similar they are to the input, by default. This feature is expected to be hitting iOS and Android some time this fall.
Branching off from Image Search, the other feature is called Find It On eBay, and it allows users to share a link from just about anywhere, including major social platforms, to the eBay app, and be presented with listings matching products found on that page. Using the same computer vision and deep learning technology as Image Search, along with a few other tweaks, Find It On eBay is built to figure out what product on a page has caught a user’s eye, find matching listings, then present them in ranked order according to how closely they match. Unlike Image Search, Find It On eBay will only be found on Android this fall, and eBay’s press release did not put out a timeline for it to come to iOS. At this time, there has been no word on either of these features coming to desktop platforms in the future, either.
