Introducing the Thread Genius API
Hello, friend! It’s been quite a while since our last post here. Our team has been hard at work this summer and today we’re excited to announce the launch of our latest product, the Thread Genius API.
Thread Genius started with a question: “How do people wear this?” (A few of you may recall, it was actually “How do people wear this shirt?”) Motivated by this question, we set out to build a new kind of visual search engine, one with which you can search style inspiration photos by the products contained within them.
Since releasing our mobile app last year, we received emails from plenty of e-commerce companies around the world looking to use our technology to build computer vision-aided features. Throughout this process, we discovered all sorts of use cases for our visual search engine Robo Billy, ranging from the obvious ones like clothing image classification and identifying products within social media photos, to less obvious ones like searching for fashion influencers and ad targeting.
We realized there were likely way more applications that could be built on top of Robo Billy than we could even fathom. For this reason, we decided to open up our visual search engine and enable developers from all over to build cool, creative applications with it.
We’ve designed our API around a few simple building blocks. Let’s go over them.
Public + Private Catalogs
At the heart of our API is the catalog, a collection of images indexed by our feature extraction neural network. Behind the scenes, each image is converted into a vector and placed into a search tree optimized for millisecond nearest-neighbor lookups. Catalogs can grow to millions of images without much quality or speed degradation.
You can build and manage your own catalogs, or even use any of the publicly available ones. The two public catalogs we currently have available are Shopstyle, an aggregation of 2M fashion products currently on sale in the US, and Bloglovin’ Fashion, images pulled from Bloglovin’, an aggregator of blog content. These catalogs are refreshed monthly with new content.
Search API
…and then comes Search. As you can guess, once you’ve picked a catalog, you can search it.
Queries come in two flavors: Image queries and predicted keyword queries. Both actually make use of our feature extractor neural net.
When adding images into a catalog, you can actually specify a crop, which tells our indexer to only index that part of the photo. This is useful for searching for things within images, like clothing products.
Prediction API
How might you figure out which crops to specify for a photo? That’s where our Prediction feature comes handy. Our Prediction API takes in a photo and predicts one of two things: where the products are (bounding box detection) and which product-based attributes are associated with the photo (tagging).
You can also send in multiple images in batch — they’ll just be queued up and you can check on their predictions later. On free tier, you can send up to 10,000 prediction queries per month.
Dashboard
We’ve created a barebones UI where you can find your API key and check up on your API usage. It’s still very much a work in progress and we’re definitely looking to take suggestions for what to include here.
Getting started
Getting started with our API is simple and free. Sign up and grab an API key while they’re still hot: threadgenius.co/signup.
And some helpful links from here:
Share some love on Product Hunt 💝
If you got this far and liked what you’ve seen, please show your support by upvoting our Product Hunt post. We are on the front page today!
