app.py

import gradio as gr
from PIL import Image
import os

def get_identity_images(path="images/identities"):
    ims = Image.open([os.path.join(path,im) for im in os.listdir(path)])
    return ims

with gr.Blocks() as demo:
    gr.Markdown("""
                ## Stable Bias:  Analyzing Societal Representations in Diffusion Models
            """)
    gr.HTML('''
     <p style="margin-bottom: 10px; font-size: 94%">This is the demo page for the "Stable Bias" paper, which aims to explore and quantify social biases in text-to-image systems. <br> This work was done by <a href='https://huggingface.co/sasha' style='text-decoration: underline;' target='_blank'> Alexandra Sasha Luccioni (Hugging Face) </a>, <a href='https://huggingface.co/cakiki' style='text-decoration: underline;' target='_blank'> Christopher Akiki (ScaDS.AI, Leipzig University)</a>, <a href='https://huggingface.co/meg' style='text-decoration: underline;' target='_blank'> Margaret Mitchell (Hugging Face) </a> and  <a href='https://huggingface.co/yjernite' style='text-decoration: underline;' target='_blank'> Yacine Jernite (Hugging Face) </a> .</p>
          ''')
    
    gr.HTML('''
     <p style="margin-bottom: 14px; font-size: 100%">  As AI-enabled Text-to-Image systems are becoming increasingly used, characterizing the social biases they exhibit is a necessary first step to lowering their risk of discriminatory outcomes. <br> We propose a new method for exploring and quantifying social biases in these kinds of systems by directly comparing collections of generated images designed to showcase a system’s variation across social attributes — gender and ethnicity — and target attributes for bias evaluation — professions and gender-coded adjectives. <br> We compare three models: Stable Diffusion v.1.4, Stable Diffusion v.2., and Dall-E 2, and present some of our key findings below:</p>
          ''')    

    with gr.Accordion("Identity group results (ethnicity and gender)", open=False):
         gr.HTML('''
         <p style="margin-bottom: 14px; font-size: 100%">  One of the approaches that we adopted in our work is hierarchical clustering of the images generated by the text-to-image systems in response to prompts that include identity terms with regards to ethnicity and gender. <br> We computed 3 different numbers of clusters (12, 24 and 48) and created an <a href='https://huggingface.co/spaces/society-ethics/DiffusionFaceClustering' style='text-decoration: underline;' target='_blank'> Identity Representation Demo </a> that allows for the exploration of the different clusters and their contents. </p>
              ''')    
         with gr.Row():
            impath = "images/identities"
            identity_gallery = gr.Gallery([os.path.join(impath,im) for im in os.listdir(impath)],
            label="Identity cluster images", show_label=False, elem_id="gallery"
        ).style(grid=3, height="auto")
         gr.HTML('''
         <p style="margin-bottom: 14px; font-size: 100%"> TO DO: talk about what we see above. <br> Continue exploring the demo on your own to uncover other patterns!  </p>
              ''')    

    with gr.Accordion("Bias Exploration", open=False):
         gr.HTML('''
         <p style="margin-bottom: 14px; font-size: 100%">  We queried our 3 systems with prompts that included names of professions, and one of our goals was to explore the social biases of these models. <br> Since artificial depictions of fictive
humans have no inherent gender or ethnicity nor do they belong to socially-constructed groups, we pursued our analysis <b> without </b> ascribing gender and ethnicity categories to the images generated. <b> We do this by calculating the correlations between the professions and the different identity clusters that we identified. <br> Using both the <a href='https://huggingface.co/spaces/society-ethics/DiffusionClustering' style='text-decoration: underline;' target='_blank'> Diffusion Cluster Explorer </a> and the <a href='https://huggingface.co/spaces/society-ethics/DiffusionFaceClustering' style='text-decoration: underline;' target='_blank'> Identity Representation Demo </a>, we can see which clusters are most correlated with each profession and what identities are in these clusters.</p>
              ''')    
         with gr.Row():
             gr.HTML('''
         <p style="margin-bottom: 14px; font-size: 100%">  Using the <a href='https://huggingface.co/spaces/society-ethics/DiffusionClustering' style='text-decoration: underline;' target='_blank'> Diffusion Cluster Explorer </a>, we can see that the top cluster for the CEO and director professions is Cluster 4: </p> ''')    
             ceo_img = gr.Image(Image.open("images/bias/ceo_dir.png"), label = "CEO Image", show_label=False)   
                
         with gr.Row():
             gr.HTML('''
         <p style="margin-bottom: 14px; font-size: 100%"> Going back to the <a href='https://huggingface.co/spaces/society-ethics/DiffusionFaceClustering' style='text-decoration: underline;' target='_blank'> Identity Representation Demo </a>, we can see that the most represented gender term is man (56%  of the cluster) and White (29% of the cluster).  </p> ''')    
             cluster4 = gr.Image(Image.open("images/bias/Cluster4.png"), label = "Cluster 4 Image", show_label=False)             
         with gr.Row():
             gr.HTML('''
         <p style="margin-bottom: 14px; font-size: 100%">  If we look at the cluster representation of professions such as social assistant and social worker, we can observe that the former is best represented by Cluster 2, whereas the latter has a more uniform representation across multiple clusters:   </p> ''')    
             social_img = gr.Image(Image.open("images/bias/social.png"), label = "social image", show_label=False)   
                
         with gr.Row():
             gr.HTML('''
         <p style="margin-bottom: 14px; font-size: 100%"> Cluster 2 is best represented by the gender term is woman (81%) as well as Latinx (19%). </p> ''')    
             cluster4 = gr.Image(Image.open("images/bias/Cluster2.png"), label = "Cluster 2 Image", show_label=False)  
         with gr.Row():
             gr.HTML('''
             <p style="margin-bottom: 14px; font-size: 100%"> TO DO: talk about what we see above. <br> Continue exploring the demo on your own to uncover other patterns!  </p>''')
                     
demo.launch(debug=True)