Spaces:
Runtime error
Runtime error
| 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) | |