Google's Mood Board Search Trains a Computer to Recognize Concepts

Google presented Mood Board Search – a new machine learning-powered research tool that lets you train a computer to recognize visual concepts using mood boards. The search is useful for abstract prompts, like "peaceful" or even "vibrant color palette that feels part memory, part dream."

The researchers' goal is "to design a flexible and approachable interface so people without ML expertise can train a computer to recognize a visual concept as they see it." 

All you need to do is upload images that evoke a certain style, mood, or “feel” of an idea, select an image set to search, train your Concept Activation Vector (CAV), and explore the results to see how well the model expresses your concept. 

Here's how it works:

You can also signal which images are more important to a visual concept by upweighting or downweighting images. Then, you can use the Focus mode to understand which part of an image best matches the visual concept from the AI's perspective.

Find out more about Mood Board Search here and get the code on GitHub.