paint-brush
Holistic Evaluation of Text-to-Image Models: Human evaluation procedureby@autoencoder
110 reads

Holistic Evaluation of Text-to-Image Models: Human evaluation procedure

tldt arrow

Too Long; Didn't Read

This section covers the process of obtaining human feedback on AI-generated images using Amazon Mechanical Turk. The project required annotators to meet specific qualifications and paid workers based on an hourly wage. The survey and annotation process involved compliance with Institutional Review Board (IRB) standards.
featured image - Holistic Evaluation of Text-to-Image Models: Human evaluation procedure
Auto Encoder: How to Ignore the Signal Noise HackerNoon profile picture

Authors:

(1) Tony Lee, Stanford with Equal contribution;

(2) Michihiro Yasunaga, Stanford with Equal contribution;

(3) Chenlin Meng, Stanford with Equal contribution;

(4) Yifan Mai, Stanford;

(5) Joon Sung Park, Stanford;

(6) Agrim Gupta, Stanford;

(7) Yunzhi Zhang, Stanford;

(8) Deepak Narayanan, Microsoft;

(9) Hannah Benita Teufel, Aleph Alpha;

(10) Marco Bellagente, Aleph Alpha;

(11) Minguk Kang, POSTECH;

(12) Taesung Park, Adobe;

(13) Jure Leskovec, Stanford;

(14) Jun-Yan Zhu, CMU;

(15) Li Fei-Fei, Stanford;

(16) Jiajun Wu, Stanford;

(17) Stefano Ermon, Stanford;

(18) Percy Liang, Stanford.

Abstract and 1 Introduction

2 Core framework

3 Aspects

4 Scenarios

5 Metrics

6 Models

7 Experiments and results

8 Related work

9 Conclusion

10 Limitations

Author contributions, Acknowledgments and References

A Datasheet

B Scenario details

C Metric details

D Model details

E Human evaluation procedure

E Human evaluation procedure

E.1 Amazon Mechanical Turk

We used the Amazon Mechanical Turk (MTurk) platform to receive human feedback on the AIgenerated images. Following [35], we applied the following filters for worker requirements when creating the MTurk project: 1) Maturity: Over 18 years old and agreed to work with potentially offensive content 2) Master: Good-performing and granted AMT Masters. We required five different annotators per sample. Figure 6 shows the design layout of the survey.


Based on an hourly wage of $16 per hour, each annotator was paid $0.02 for answering a single multiple-choice question. The total amount spent for human annotations was $13,433.55.

E.2 Human Subjects Institutional Review Board (IRB)

Figure 6: Human annotation interface. Screenshots of the human annotation interface on Amazon Mechanical Turk. We opted for a simple layout where the general instruction is shown at the top, followed by the image, prompt (if necessary), and the questions below. Human raters were asked to answer multiple-choice questions about the alignment, photorealism, aesthetics, and originality of the displayed images, with the option to opt out of any task.


This paper is available on arxiv under CC BY 4.0 DEED license.