May 18, 2017 | Written by: Kevin Gong
Categorized: Community | Watson | What's New
Share this post:
A picture can be worth a thousand words, but only some of those words may be meaningful for a business working to solve a specific problem using images.
Since introducing the Visual Recognition API, we’ve made many updates to improve how businesses and developers can gain insights from images. We announced both price reductions for custom classifiers as well as a general tagging update for Visual Recognition that saw users reporting significantly higher accuracies for general tagging due to an active vocabulary that was 2.5 times larger than before. Those updates are part of ongoing improvements we’re making to two of our longstanding feature categories: general tagging and custom classifiers.
What we’ve learned is that domain expertise matters for data to be useful for businesses. Photos and images can be tagged with an infinite number of classifiers, but one of our goals is to provide users with enhanced accuracy for certain domains they want to look at.
We’re excited to announce today the introduction of a new category of features: pre-built custom models. These models are developed in-house at IBM, cutting down on the time-intensive task of manual classifier training and enabling users to more easily identify domain-specific content. With this new expanded offering, the Visual Recognition API expands on the traditional industry model by offering three distinct layers for users depending on their needs: out-of-the-box capabilities, pre-built industry models, and fully-custom models.
Watson Visual Recognition food model
The first pre-built custom model we’re introducing is food recognition. While Watson has already been able to detect food as part of general tagging, the new food model provides enhanced specificity and accuracy for food items, based on over 2,000 tags.
Similar to general tagging, each returned result will be accompanied by a classification score, and future updates will allow users to create custom classifiers on top of this model. The new expanded offering allows users to quickly assess the dishes within an image and turn these into actionable insights.
Who should use the food model?
We developed the food model with the restaurant, health & fitness, lifestyle, and travel industries in mind. Our inspiration came from hearing from companies that aim to replace the arduous and manual process of food logging with automatic food identification using Visual Recognition. From social media analytics to content curation, the food model makes it easier to automatically manage content and gather insights.
Restaurants and chains can use the food model to monitor the appeal of their products on social media. The food model can not only help businesses understand how frequently their different products are being shared in social media images, but also be paired with other Watson APIs like Tone Analyzer and Natural Language Understanding to understand the context and sentiment surrounding these images.
Meanwhile, the food model can also empower review and delivery businesses to automatically classify both user-submitted images and official restaurant photos. With image databases in the range of millions for businesses, automatic content tagging with the food model enables food-centric services to efficiently manage their content.
Need help or have questions?
We’re excited to see what you build with Watson Visual Recognition, and we’re happy to help you along the way. Try the food recognition feature, read the documentation, and share any questions or comments you have on our developerWorks forums.