This is a Plain English Papers summary of a research paper called AI Dataset Helps Generate More Inclusive Public Space Images, Study Shows. If you like these kinds of analysis, you should join AImodels.fyi or follow us on Twitter.
Overview
- LIVS is a dataset for aligning text-to-image AI models with values of inclusivity
- Created with input from urban planners, accessibility experts, and diverse communities
- Contains 10,000 prompt pairs showing preferred vs. non-preferred representations of public spaces
- Covers five values: accessibility, intersectionality, safety, sustainability, and inclusivity
- Demonstrates effectiveness with DPO (Direct Preference Optimization) training method
- Models fine-tuned with LIVS show improved representation of diverse groups in public spaces
Plain English Explanation
LIVS (Living In Valuable Spaces) is a new dataset created to help make AI image generation systems better at creating pictures of public spaces that work for everyone.
Think about the last time you visited a public park, street, or plaza. Was it easy to navigate if you had a ...