Research map / Image generation / Text-to-image diffusion
DALL-Eval: Probing the Reasoning Skills and Social Biases of Text-to-Image Generation Models
Jaemin Cho, Abhay Zala, Mohit Bansal
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 |
|---|---|---|---|
| compare Stable Diffusion with minDALL-E on counting and spatial skills | text-to-image models finetuned on PaintSkills · PaintSkills | accuracy (%): count 37.8 vs 47.5; spatial 7.9 vs 50.7; object 96.2 vs 89.9 | Mixed |
| model generation vs oracle upper bound on count and spatial | T2I models · PaintSkills | accuracy (%): oracle 97.8/96.2 vs best 47.5/50.7 | Hurt |
| reduce finetuning data from 100% to 50% (ablation) | minDALL-E and Stable Diffusion · PaintSkills | avg accuracy (%): 62.7 -> 64.3 (minDALL-E); 47.3 -> 48.6 (SD) | No clear effect |
| reduce finetuning data from 100% to 10% (ablation) | minDALL-E · PaintSkills spatial | accuracy (%): 50.7 -> 38.2 | Hurt |
| use BLIP-2 instead of CLIP for gender detection (ablation) | bias evaluation pipeline · COCO bias test and Adience | accuracy (%): 66% -> 82% | Improved |
| use BLIP-2 instead of CLIP for attribute detection (ablation) | bias evaluation pipeline · attribute detection | accuracy (%): 79% -> 92% | Improved |
| compare gender MAD across models | T2I models · gender-neutral prompts | MAD: minDALL-E 0.1984 vs Karlo 0.3545 vs SD 0.3618 | Mixed |
| compare skin tone MAD across models | T2I models · diagnostic prompts | MAD: 0.1687 / 0.1707 / 0.1698 | No clear effect |
Closest papers by meaning
- MMMG: A Massive, Multidisciplinary, Multi-Tier Generation Benchmark for Text-to-Image Reasoning (NeurIPS 2025 · similarity 0.89)
- ROVER: Benchmarking Reciprocal Cross-Modal Reasoning for Omnimodal Generation (ICLR 2026 · similarity 0.89)
- LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models (ICLR 2026 · similarity 0.88)
- On Fairness of Unified Multimodal Large Language Model for Image Generation (NeurIPS 2025 · similarity 0.88)
- GIR-Bench: Versatile Benchmark for Generating Images with Reasoning (ICLR 2026 · similarity 0.88)
- DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Models (ICML 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.
