Summary of Decomposed Evaluations Of Geographic Disparities in Text-to-image Models, by Abhishek Sureddy et al.
Decomposed evaluations of geographic disparities in text-to-image models
by Abhishek Sureddy, Dishant Padalia, Nandhinee Periyakaruppa, Oindrila Saha, Adina Williams, Adriana Romero-Soriano, Megan Richards, Polina Kirichenko, Melissa Hall
First submitted to arxiv on: 17 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This research paper introduces a novel set of metrics called Decomposed Indicators of Disparities in Image Generation (Decomposed-DIG) to measure geographic disparities in the depiction of objects and backgrounds in generated images. The authors apply this new metric to audit a widely used latent diffusion model, finding that generated images tend to have more realistic object representations than background representations, with larger regional disparities observed in generated backgrounds. The study uses Decomposed-DIG to identify specific examples of disparities, such as stereotypical background generation in Africa and unrealistically placing objects in outdoor settings. To address these issues, the authors propose a new prompting structure that improves worst-region performance by 52% and average background diversity by 20%. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research paper is about finding out if computer-generated images are biased towards certain regions or cultures. Right now, there’s no good way to measure this bias, so researchers have been relying on humans to evaluate the images, which takes a lot of time and money. The authors of this paper created new metrics that can measure bias in specific parts of an image, like objects or backgrounds. They used these metrics to test a popular computer program for generating images and found some surprising results. For example, the program had trouble generating realistic cars and buildings for certain regions. To fix this problem, the researchers came up with a new way to tell the program what kind of image to generate, which made a big difference. |
Keywords
» Artificial intelligence » Diffusion model » Image generation » Prompting
