How biased are text-to-image models?

Image 1: Input prompt: “ A photo of a doctor”
Image 2: Input prompt: “A photo of a nurse”
Figure 1: A text prompt is fed to the text encoder to give an output. An image is put into the image encoder to give an output. These two outputs were compared to see how similar the text is to the image.
Figure 2: The dark blue lines indicate the counts of the prompt “doctor” being associated with male images. The light blue lines indicate the counts of the prompt “doctor” being associated with female images. The x-axis represents the similarity scores between both female and male doctor images and the prompt “doctor”, and the y-axis represents the number of images.
Figure 3: The dark blue lines indicate the counts of the prompt “nurse” being associated with male images. The light blue lines indicate the counts of the prompt “nurse” being associated with female images. The x-axis represents the similarity scores between both female and male nurse images and the prompt “nurse”, and the y-axis represents the number of images.
Figure 4: This shows the production of an equal representation of Asian, Black, Hispanic, and White students when producing a 100 images of a computer science student.
Figure 5: These are the associated outputs for the inputs when it considers both race and gender.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Cornell Data Science

Cornell Data Science

Cornell Data Science is an engineering project team @Cornell that seeks to prepare students for a career in data science.