Teaching Humans Subtle Differences with DIFFusion
This addresses the challenge for domain experts and learners in fields where subtle visual distinctions are hard to articulate, offering a novel tool for visual learning and scientific research.
The paper tackles the problem of teaching humans to recognize subtle visual differences between categories by using generative models to create counterfactual visualizations with minimal discriminative features, resulting in significantly improved human differentiation of fine-grained classes across domains like black hole simulations and medical imaging.
Scientific expertise often requires recognizing subtle visual differences that remain challenging to articulate even for domain experts. We present a system that leverages generative models to automatically discover and visualize minimal discriminative features between categories while preserving instance identity. Our method generates counterfactual visualizations with subtle, targeted transformations between classes, performing well even in domains where data is sparse, examples are unpaired, and category boundaries resist verbal description. Experiments across six domains, including black hole simulations, butterfly taxonomy, and medical imaging, demonstrate accurate transitions with limited training data, highlighting both established discriminative features and novel subtle distinctions that measurably improved category differentiation. User studies confirm our generated counterfactuals significantly outperform traditional approaches in teaching humans to correctly differentiate between fine-grained classes, showing the potential of generative models to advance visual learning and scientific research.