Research map / Image generation / Text-to-image diffusion
CuRe: Cultural Gaps in the Long Tail of Text-to-Image Systems
Aniket Rege, Zinnia Nie, Mahesh Ramesh, Unmesh Raskar, Zhuoran Yu, Aditya Kusupati, Yong Jae Lee, Ramya Korlakai Vinayak
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use category prompt P(c) instead of name prompt P(n) (ablation) | sim(I(n),P) ITA scorer · CuRe, FLUX.1 [dev] ITA vs PS | Spearman rho: 0.38 -> 0.34 | Hurt |
| use region-only prompt P(r) instead of name prompt P(n) (ablation) | sim(I(n),P) ITA scorer · CuRe, FLUX.1 [dev] ITA vs PS | Spearman rho: 0.38 -> 0.12 | Hurt |
| combine category and region prompts P({c,r}) in the ITA scorer (ablation) | phi_ITA · CuRe, FLUX.1 [dev] ITA vs PS | Spearman rho: 0.40 -> 0.42 | Improved |
| use marginal-utility ITA scorer instead of PickScore | phi_ITA({c,r}) · CuRe, FLUX.1 [dev] | Spearman rho vs phi*_PS: 0.34 -> 0.42 | Improved |
| use phi_PS(n) with no ground truth instead of phi_GT(n) | SigLIP 2 PS scorer · CuRe, FLUX.1 [dev] | Spearman rho vs phi*_PS: 0.44 -> 0.32 | Hurt |
| use divergence PS scorer on SD 1.5 instead of FLUX.1 | delta phi_PS({n,c}) · CuRe SigLIP 2 | Spearman rho vs phi*_PS: -0.31 -> 0.04 | Hurt |
| use MIA-based phi_DIV instead of LPIPS(n) | diversity scorer · CuRe, FLUX.1 [dev] | Spearman rho vs phi*_CuRe: -0.11 -> -0.20 | Improved |
| use phi_DIV instead of Vendi score VS(c) | diversity scorer · CuRe, SD 3.5 Large | Spearman rho vs phi*_CuRe: 0.03 -> -0.22 | Improved |
| add quality weighting to Vendi score (qVS) (ablation) | VS(c) · CuRe, FLUX.1 [dev] | Spearman rho vs phi*_CuRe: -0.02 -> -0.02 | No clear effect |
| use SigLIP 2 instead of DINOv2 embeddings for PS benchmark scoring vs ELO | phi_PS divergence · Image Arena ELO | Spearman rho with ELO: 0.564 vs -0.100 | Mixed |
| use OpenAI WIT CLIP backbone for ITA scorer vs ELO | phi_ITA · Image Arena ELO | Spearman rho with ELO: -0.829 | Hurt |
| use Gemini 2.0 Flash judge instead of SigLIP 2 phi_GT on FLUX.1 | Gemini 2.0 Flash PS scorer · CuRe, FLUX.1 [dev] | Spearman rho vs phi*_PS: 0.40 vs 0.44 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ImageReward — ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation (NeurIPS 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Beyond Aesthetics: Cultural Competence in Text-to-Image Models (NeurIPS 2024 · similarity 0.93)
- No Filter: Cultural and Socioeconomic Diversity in Contrastive Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes (ICLR 2025 · similarity 0.87)
- Culture In a Frame: C$^3$B as a Comic-Based Benchmark for Multimodal Culturally Awareness (ICLR 2026 · similarity 0.87)
- Multilingual Diversity Improves Vision-Language Representations (NeurIPS 2024 · similarity 0.87)
- All Languages Matter: Evaluating LMMs on Culturally Diverse 100 Languages (CVPR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
