ConfusionLens: Dynamic and Interactive Visualization for Performance Analysis of Multiclass Image Classifiers

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Building higher-quality image classification models requires better performance analysis (PA) methods to help understand their behaviors. We propose ConfusionLens, a dynamic and interactive visualization interface that augments a conventional confusion matrix with focus+context visualization. This interface makes it possible to adaptively provide relevant information for different kinds of PA tasks. Specifically, it allows users to seamlessly switch table layouts among three views (overall view, class-level view, and between-class view) while observing all of the instance images in a single screen. This paper presents a ConfusionLens prototype that supports hundreds of instances and its several extensions to further support practical PA tasks, such as activation map visualization and instance sorting/filtering.

Original languageEnglish
Title of host publicationUIST 2022 Adjunct - Adjunct Proceedings of the 35th Annual ACM Symposium on User Interface Software and Technology
PublisherAssociation for Computing Machinery, Inc
ISBN (Electronic)9781450393218
DOIs
Publication statusPublished - 2022 Oct 29
Event35th Annual ACM Symposium on User Interface Software and Technology, UIST 2022 - Bend, United States
Duration: 2022 Oct 292022 Nov 2

Publication series

NameUIST 2022 Adjunct - Adjunct Proceedings of the 35th Annual ACM Symposium on User Interface Software and Technology

Conference

Conference35th Annual ACM Symposium on User Interface Software and Technology, UIST 2022
Country/TerritoryUnited States
CityBend
Period22/10/2922/11/2

Keywords

  • Confusion Matrices
  • Image Classification
  • Interactive Visualization
  • Performance Analysis
  • User Interface

ASJC Scopus subject areas

  • Software
  • Computer Graphics and Computer-Aided Design
  • Human-Computer Interaction

Fingerprint

Dive into the research topics of 'ConfusionLens: Dynamic and Interactive Visualization for Performance Analysis of Multiclass Image Classifiers'. Together they form a unique fingerprint.

Cite this