Pinterest is experimenting with new visual search technology. The company asks us to imagine discovering new products and ideas on Pinterest based on visually similar objects, as Pinned by people with similar tastes.
We’re told that’s the idea behind some of the latest technology the company’s Visual Discovery team is working on. Pinterest is sharing details on new experiments on its blog and in a newly released white paper by Kevin Jing, its head of Visual Discovery and founder of VisualGraph, which Pinterest acquired in early 2014.
“After the acquisition of VisualGraph last year, the team began work on a large-scale, cost-effective machine vision pipeline and stack to improve visual search results and the 1.5 trillion recommendations we serve each year,” a Pinterest spokesperson tells WebProNews. “Visual signals are used in almost everything we do, and with more than 50 billion Pins in the system, we have one of the largest, unique and richly annotated datasets available.”
Pinterest is experimenting with real-time related pin recommendations and object recognition. The former, for new pins entering the system (which haven’t had related pins in the past), are based on visual similarities of other pins in the system. According to the company, over 90% of pins now have related pins served up as recommendations.
The object recognition can power recommendations based on specific objects in a pin’s image. The company says it’s experimenting with blending these pins into related pin recommendations below a pin. A pin’s image of an athlete wearing a backpack, for example, might lead to related pins for a similar backpack, running shoes and/or athletic wear.
The company plans to roll out a version of this later this year depending on how the experimentation goes.
“By sharing our machine vision pipeline implementation details and the experiences of launching products (detailed in the white paper), we hope visual discovery can be more widely incorporated into today’s consumer apps, for greater use by everyday people,” the spokesperson says.
Read the white paper here and find more info in this post.
Images via Pinterest