Innovating Interior Business using Computer Vision and Visual Search of the Furniture
The furniture market is worth a trillion dollars and is the second largest offline retailing business sector in the world after alcoholic beverages. EU based startup with millions of Venture Capital invested was planning to make buying furniture online more accessible and convenient. But their existing system was not fast enough to provide good consumer experience. For that purpose they approached Vostok AI team.
Challenges with Existing Solution
When customer tried to implement visual search on their website, they came across some of the challenges:
Searching for particular product take between 15 to 20 minutes;
Most of the products can’t be found by text description;
Product description & search were subjective;
As a result, most of the consumers couldn’t find the product and need to leave.
Solution.
Based on database of over 1 million images we trained our neural network to find similarities between objects and styles (including items of the different category).
Example: “a person liked a table in a cafe, took a picture of it and upload it to the visual search engine and within seconds got
Results.
Whole project was created within 4 months to develop visual search for the application. The application (using neural networks) were able to recognize:
Object’s shape;
Color
Texture
And context of object use (to match existing object with other objects that match overall style)
As a result of the project, the company got state-of-the-art visual search engine and was able to raise additional investments (1 mln. Euro) from EU-based VC funds.
Customer quote: “Using this solution we will be able to automatically categorize newly added products on the platform and create descriptions for them. Designers will be able to upload 3D rendering images onto the Design Scan platform that will recognize items in the interior and provide a choice of similar items that are available in stock. And last but not least, neural networks will make life easier for our end-users, because they will be able to upload photos and the Design Scan platform will find the exact item and lead to the seller“