Human Evaluation on Interpretation by GuessWrongLabel

GuessWrongLabel User Interface (UI) was developed to support the paper:

• How Useful Are the Machine-Generated Interpretations to General Users?A Human Evaluation on Guessing the Wrongly Predicted Labels. Hua Shen, Ting-Hao (Kenneth) Huang. The 8th AAAI Conference on Human Computation and Crowdsourcing (HCOMP), 2020.

Usage

User Interface for Evaluating Interpretation by General User:

User Interface of two experimental conditions, including [Interpretation] and [No-interpretation], are both provided in this repo with six examples.

• UI for [No-Interpretation] condition is in "/No_Interpretation_Interface/" with original images in "/condition1/";
• UI for [Interpretation] condition is in "/Interpretation_Interface/" with images and saliency maps in "/condition2/".

User Interface Demo:

Machine-generated Interpretations:

The machine-generated interpretation codes using Pytorch is in "/machine_generated_interpretation/".

Full paper access

The full paper can be found here.