0 (1)

Visual Search at Zalando, ASOS

The internet was close to breaking when ASOS announced its visual search feature – and for good reason: The fact that the fashion retail giant has made “search by image” a permanent tool for shoppers on its mobile app is further testament to the fact that images are indeed the future of search, as Pinterest CEO Ben Silbermann predicted just a few months ago. ASOS also joins other retail giants like Zalando, who have offered visual search since 2016.

Although visual search is the hot topic of the moment, not many people are talking about how it works. To give you some insights into the revolutionary technology that powers visual search tools like at Zalando and ASOS, we at Fashwell have put together a guide on what visual search actually is, how it works and why it’s the future of online fashion retail.

Why Visual Search?

To explain why using images makes so much sense for search, it’s important to know how the human brain works. Ninety percent of the information transferred to the brain is visual – and we process images 60,000 times faster than text-based information. It’s no wonder, then, that images and videos are quickly turning into our primary form of communication. It’s a lot faster and it better communicates what we are trying to say. Especially in fashion it’s difficult to put the style of a product into words.

Why Visual Search Makes Sense for Fashion

Shoppers are inspired by images. When they scroll through Instagram or click through a blog, it’s the picture of the pink stilettos or the black silk top that catches their eye. It only makes sense then that when shopping for apparel – which is also the fastest growing sector in eCommerce – that the communication between retailer and customer is also image-based. That is, through visual search.

Zalando Asos visual search app fashion ecommerce
Zalando visual search page vs. ASOS visual search

Zalando and ASOS have recognized that most of their shoppers face a common dilemma: A shopper, let’s call her Ashley, sees a picture of Rihanna on holidays on her Instagram feed. In this picture, she’s wearing a white V-neck jumpsuit. Ashley loves Rihanna’s style, so she wants to buy the same jumpsuit. She visits her favorite online fashion retailer and types “white jumpsuit” into the search bar. She scrolls through the results, but they aren’t good. Same thing on a second site. When she clicks through to the third site, Ashley is frustrated and no longer in the mood to shop. The result: no jumpsuit for Ashley, and no sale for the retailers.

This would never have happened with visual search, because it’s basically like typing “white minimalist jumpsuit that Rihanna wore in Portofino” into the search bar, but actually getting viable results. But all with one image.

How Visual Search Actually Works

Deep learning technology makes image recognition and visual search possible. Basically, visual search is the localization of products within an image and the placement of those products within the greater context of a shop catalog. To achieve this, a fashion universe, also known as a feature space, must be built. This is a collection of all the products sold by a brand or retailer. The fashion universe has one simple feature: products of a similar style are close together, products of different styles are far apart.

fashion visual search deep learning
A neural network that is inspired by the structure of our brain creates the fashion universe. It consists of many different layers of neurons, each with a specific function. As they comb through the images, the neurons light up once they hit upon a specific visual feature.

Which leads us to the second part of visual search: The input image is analyzed, the products are localized. Then, they’re mapped within the feature space. In our example, we input the image of Rihanna, the products – jumpsuit, sunglasses, sandals – are localized. They are then grouped with similar products in a feature space, like those of Zalando or ASOS.

In order to achieve highly accurate image data analysis results, the machine requires massive amounts of training data. It must learn from other fashion images not only what the difference between a dress and a jumpsuit is, but also between two jumpsuits, like a catsuit or a romper.

Contrary to popular belief, visual search tools are actually super easy to implement

A visual search tool, like that offered by Fashwell, is easily integrable into any mobile or desktop eCommerce platform. We work with retailers, like Zalando, to turn their eCommerce channels into the best possible shopping experience for their customers. Visual search checks off all the expectations of modern shoppers: choice, ease, personalization and curation. Brands and eCommerce retailers who offer shop by image can count on increased conversions, higher revenue and customer retention.