ScatterUQ Demo

This page is a demo for the paper ScatterUQ: Interactive Uncertainty Visualizations for Multiclass Deep Learning Problems.


  

Normally the user can use the sliders below to adjust the outlier tolerance and class confidence threshold. When the sliders are changed, the use case that each inference sample falls under may change. However, since this is a static demo, these values are fixed in place.

Outlier Tolerance:    95%
 

Class Confidence Threshold:    70%

Number of Samples in Each Label  

T-shirt/top Trouser Pullover Dress Coat Sandal Shirt Sneaker Bag Ankle boot CLASS_CONFUSION OOD
Loading...
Filter Labels
Select All  /   Deselect All

Scatter UQ Inference Samples

There is no data to show

Global Scatterplot with UMAP