The best thing about computer vision models is you can teach it to see almost anything, exactly the way you want it to see those concepts. Here are 3 unique concepts we’ve built custom models for:
With over 400 million stories posted on their platform, Wattpad’s tagline “where stories live” is apt. Since its founding in 2006, it has grown into a diverse community of more than 65 million readers, authors, and artists sharing and viewing both written and visual works of art. While the site’s content guidelines prohibit the posting of certain types of written and visual content, like any site that relies on user-generated content (UGC,) the risk of rules not being followed is always high. As such, the company relied on its community team to moderate this content. Unfortunately, a surge in the volume of uploaded content saw this spending most of their days away from the job they enjoyed most and was of the most value to Wattpad, i.e., help to build the community. So, like many companies before them, Wattpad turned to computer vision.
While our NSFW model is widely used to recognize concepts like nudity and violence, its training data consisted of still life photographs. As Wattpad’s visual content primarily consisted of illustrations or drawings, they needed a distinct model. With data from their actual platform, our DST team was able to build them a custom moderation model that could recognize the types of photos they actually received, i.e., drawings and still life images, in just a few weeks. This model allows for far more efficiency in their moderation efforts and lets their community team focus on the user-oriented tasks that are far more advantageous for the company and its users.
While our general model can recognize over 11,000 concepts, including live animals, for one of our customers, an online marketplace, they needed it to recognize something a little more grim: animal parts. Why? Well, while online marketplaces provide their users with a secure platform on which they can buy and sell almost everything: clothes, cars, used furniture, the platform strictly prohibits the sale of body parts, like skin and bones, from protected species, like gopher tortoise shells or tiger skin. It’s a crucial limitation. The trading of these commodities doesn’t just harm the animals, it harms humans, both economically and physically.
Much more needs to be done to stem the flow of the wildlife trade, but with this custom model, the company is at least on the right side of this problem. By preventing purveyors of such items from utilizing their platform, they are making it harder for them to find buyers, at least among the general public, and so less commercially viable.
Acne (& other skin problems)
The beauty industry is big business. Every day, consumers around the world make purchases on their quest for flawless skin. Still, it can be difficult to figure out which product will work best for their concerns from the description alone.
One of our customers wanted to help buyers in their purchasing decisions, and so, turned to computer vision. With just a few images, our team was able to train a custom model that recognized a wide array of skin-related matters, like uneven skin tone or enlarged pores. When integrated with their product-suggestion platform, the model’s predictions allow for only the most relevant products to be recommended, even advising the users on what products to use when, and even where on the face the customer should pay particular attention, based on the detected concerns.
With the ability to train models to recognize concepts that are unique to you, computer vision models are as customizable a tool as they come. Sign up for a free API account and see what you can train a model to see and recognize today.