A new 60-prompt benchmark shows most text-to-image models under-represent demographic groups, while LLM-guided diversification methods improve diversity without overcorrecting contextually specified attributes.
Llama 3 model card.https://github
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Beyond Overcorrection: Evaluating Diversity in T2I Models with DivBench
A new 60-prompt benchmark shows most text-to-image models under-represent demographic groups, while LLM-guided diversification methods improve diversity without overcorrecting contextually specified attributes.