Presentation
Discriminator: Exploring Highlighting Techniques in Vast Natural Image Databases
DescriptionThe number of available natural images for machine learning is staggering. Programs like Microsoft’s AI For Earth “MegaDetector” can pre-filter large datasets by reliably detecting the presence of wildlife, humans, or vehicles and determine bounding boxes for targets. Outputs can then guide further classification.
Often human observers are providing the ground truths which are then used as input for machine learning algorithms. Given the large number of image classifications humans might be involved in, optimization of the visual presentation of target AOIs is paramount. To explore this design space we developed Discriminator, an interactive webapp that allows users to dynamically filter images based on MegaDetector output and parametrically explore different presentation techniques. By directly applying image manipulation to random selections of natural images we can test visualization strategies on the fly. As an extensible tool, Discriminator can provide additional services, including gathering online image ratings, asset sorting, and image manipulation.
Often human observers are providing the ground truths which are then used as input for machine learning algorithms. Given the large number of image classifications humans might be involved in, optimization of the visual presentation of target AOIs is paramount. To explore this design space we developed Discriminator, an interactive webapp that allows users to dynamically filter images based on MegaDetector output and parametrically explore different presentation techniques. By directly applying image manipulation to random selections of natural images we can test visualization strategies on the fly. As an extensible tool, Discriminator can provide additional services, including gathering online image ratings, asset sorting, and image manipulation.
Event Type
Demonstration
TimeFriday, September 13th8:30am - 9:30am MST
LocationFLW Salon J
Perception and Performance