NAMESAKES is a benchmark for studying identity memorization in text-to-image (T2I) models — whether a model reproduces a real person's likeness from their name, or fabricates a plausible face.
It accompanies the paper NAMESAKES: Probing Identity Memorization in Text-to-Image Models.
📄 Paper: [arXiv link]
🌐 Project page:
https://namesakes-web.github.ioNAMESAKES contains ~1.2K public figures sourced from English Wikipedia, spanning a… See the full description on the dataset page:
https://huggingface.co/datasets/malper/namesakes.
