Research map / Image generation / Text-to-image diffusion
Culture in Action: Evaluating Text-to-Image Models through Social Activities
Sina Malakouti, Boqing Gong, Adriana Kovashka
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 |
|---|---|---|---|
| add hallucination (HAL) to alignment in the composite metric (ablation) | FAITH with InternVL3 · GT-FAITH correlation (all) | Spearman: 0.41 -> 0.44 | Improved |
| add exaggeration (EXAG) to ALIGN+HAL to form FAITH (ablation) | FAITH with InternVL3 · GT-FAITH correlation (all) | Spearman: 0.44 -> 0.47 | Improved |
| add exaggeration (EXAG) to ALIGN+HAL to form FAITH (ablation) | FAITH with Qwen2.5-VL · GT-FAITH correlation (all) | Spearman: 0.39 -> 0.42 | Improved |
| add HAL to ALIGN (ablation) | FAITH with Qwen2.5-VL · GT-FAITH correlation (all) | Spearman: 0.39 -> 0.39 | No clear effect |
| use descriptor-based FAITH instead of direct MLLM-as-judge | Qwen2.5-VL · GT-FAITH correlation (all) | Spearman: 0.10 -> 0.42 | Improved |
| use descriptor-based FAITH instead of direct MLLM-as-judge | InternVL3 · GT-FAITH correlation (all) | Spearman: 0.20 -> 0.47 | Improved |
| use Proposer-Refiner reference descriptor generation instead of a single Proposer (GPT-4o) (ablation) | reference descriptors · descriptor quality | Spearman: 0.28 -> 0.33 | Improved |
| use Proposer-Refiner instead of Gemini 2.5-Flash Proposer alone (ablation) | reference descriptors · descriptor quality | Spearman: 0.30 -> 0.33 | Improved |
| raise similarity threshold tau calibrated on real images from 0.29 to 0.52 (ablation) | ALIGN/HAL · correlation with humans | Spearman: 0.21 -> 0.33 | Improved |
| evaluate prompts of Global South instead of Global North countries | Qwen-Image · CULTIVate | FAITH: 0.60 -> 0.55 | Hurt |
| evaluate Global South instead of Global North countries | Qwen-Image · CULTIVate | ALIGN: 0.36 -> 0.30 | Hurt |
| evaluate Global South instead of Global North countries | Qwen-Image · CULTIVate | HAL: 0.51 -> 0.56 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CuRe — CURE: Concept Unlearning via Orthogonal Representation Editing in Diffusion Models (NeurIPS 2025) · better on 1 of 1 shared result
- 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 0.94)
- All Languages Matter: Evaluating LMMs on Culturally Diverse 100 Languages (CVPR 2025 · similarity 0.89)
- Culture In a Frame: C$^3$B as a Comic-Based Benchmark for Multimodal Culturally Awareness (ICLR 2026 · similarity 0.88)
- Culture Affordance Atlas: Reconciling Object Diversity Through Functional Mapping (AAAI 2026 · similarity 0.88)
- No Filter: Cultural and Socioeconomic Diversity in Contrastive Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- World in a Frame: Understanding Culture Mixing as a New Challenge for Vision-Language Models (CVPR 2026 · similarity 0.88)
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.
