Datasets and Prompts for Bias Analysis in Text-to-Images Models

This dataset accompanies the paper and includes the following resources:

data

Bias_Text_to_Image_Data_Dreamshape.xlsx [Spreadsheet with annotations for each generated image_Model: Dreamshape]

Bias_Text_to_Image_Data_Playground.xlsx [Spreadsheet with annotations for each generated image_Model: Playground]

Bias_Text_to_Image_Data_Proteus.xlsx [Spreadsheet with annotations for each generated image_Model: Proteus]

Bias_Text_to_Image_Data_Protogen.xlsx [Spreadsheet with annotations for each generated image_Model: Protogen]

Bias_Text_to_Image_Data_SDXL_Lightning.xlsx [Spreadsheet with annotations for each generated image_Model: SDXL]

Bias_Text_to_Image_Data_Stable_Diffusion.xlsx [Spreadsheet with annotations for each generated image_Model: SD]

prompts-list.xlsx [List of all biased prompts used for image generation]

scripts

run_summary_figures.py [Uses data to generate summary Figures 2, 3, 5, and 7 in paper]

run_appendix_figures.py [Uses data to generate all model vs. gender/skin tone ratios per subject qualifier group figures in Appendices A and B]

RQ1_Statistical_Analysis.py [Generates statistical analyses for gender polarity]

RQ2_Statistical_Analysis.py [Generates statistical analyses for skin tone polarity]

RQ3_Statistical_Analysis.py [Generates statistical analyses for model bias]

summary_figures_out [Output for summary Figures 2, 3, 5, and 7]

appendix_outputs [Output for Appendice figures]

appendix_gender_bias

Positive

Negative

appendix_skin_bias

Positive

Negative

generated-images [Archive containing all generated image files by model]

Dreamshaper

Playground

Proteus

Protogen

SDXL-Lightning

Stable_Diffusion

statistical analyses [Excel file outputs of statistical analyses for each research question]

RQ1_gender_polarity_stats_results.xlsx

RQ2_skin_tone_results.xlsx

RQ3_model_bias_stats_results.xlsx