Summary of Improving Geo-diversity Of Generated Images with Contextualized Vendi Score Guidance, by Reyhane Askari Hemmat et al.
Improving Geo-diversity of Generated Images with Contextualized Vendi Score Guidance
by Reyhane Askari Hemmat, Melissa Hall, Alicia Sun, Candace Ross, Michal Drozdzal, Adriana Romero-Soriano
First submitted to arxiv on: 6 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); 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 proposes a novel approach to increasing the diversity of generated images from text-to-image models, specifically targeting the representation of everyday objects and geographic regions. The authors introduce an inference time intervention called contextualized Vendi Score Guidance (c-VSG), which guides latent diffusion models to generate more diverse images by referencing a “memory bank” of previously generated images and real-world exemplars. The paper evaluates c-VSG using two geographically representative datasets, finding significant improvements in diversity while maintaining or improving image quality. Qualitative analyses also reveal reduced region portrayals. This work contributes to the development of text-to-image models that better reflect the geographic diversity of the world. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research paper tries to make text-to-image models more realistic by showing different types of objects and places from around the world. Right now, these models often get stuck in a rut and can’t show everyday things or certain places accurately. The authors came up with a new way to help these models be more diverse and accurate. They tested it on two sets of images that represent different parts of the world and found that it worked really well. This means we might see better, more realistic pictures in the future. |
Keywords
* Artificial intelligence * Inference
