Research map / Image generation / Text-to-image diffusion
GenExam: A Multidisciplinary Text-to-Image Exam
Zhaokai Wang, Penghao Yin, Xiangyu Zhao, Changyao Tian, Yu Qiao, Wenhai Wang, Jifeng Dai, Gen Luo
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 ground-truth image as evaluator input (ablation) | GPT-5 low evaluator · GenExam evaluator ablation | scoring point accuracy (%): 86.2 -> 88.5 | Improved |
| give ground-truth text caption instead of image (ablation) | GPT-5 low evaluator · GenExam evaluator ablation | scoring point accuracy (%): 88.5 -> 87.8 | Hurt |
| add text caption on top of ground-truth image (ablation) | GPT-5 low evaluator · GenExam evaluator ablation | scoring point accuracy (%): 88.5 -> 88.5 | No clear effect |
| raise reasoning effort from minimal to low (ablation) | GPT-5 evaluator · GenExam evaluator ablation | scoring point accuracy (%): 84.2 -> 88.5 | Improved |
| raise reasoning effort from low to medium (ablation) | GPT-5 evaluator · GenExam evaluator ablation | scoring point accuracy (%): 88.5 -> 86.7 | Hurt |
| raise reasoning effort from low to high (ablation) | GPT-5 evaluator · GenExam evaluator ablation | scoring point accuracy (%): 88.5 -> 87.6 | Hurt |
| raise reasoning effort from minimal to low (ablation) | GPT-5-mini evaluator · GenExam evaluator ablation | scoring point accuracy (%): 79.0 -> 84.3 | Improved |
| use GPT-5 low instead of GPT-4o as evaluator (ablation) | evaluator choice · GenExam evaluator ablation | scoring point accuracy (%): 68.5 -> 88.5 | Improved |
| enable thinking mode | BAGEL · GenExam overall | relaxed score (%): 11.4 -> 12.9 | Improved |
| use dedicated T2I model instead of unified MLLM | FLUX.2 dev vs BAGEL · GenExam overall | relaxed score (%): 11.4 -> 42.3 | Improved |
| closed-source instead of open-source model | Nano Banana Pro vs FLUX.2 dev · GenExam overall | strict score (%): 2.4 -> 70.2 | Improved |
Closest papers by meaning
- MMMG: A Massive, Multidisciplinary, Multi-Tier Generation Benchmark for Text-to-Image Reasoning (NeurIPS 2025 · similarity 0.90)
- When Visualizing is the First Step to Reasoning: MIRA, a Benchmark for Visual Chain-of-Thought (CVPR 2026 · similarity 0.89)
- ROVER: Benchmarking Reciprocal Cross-Modal Reasoning for Omnimodal Generation (ICLR 2026 · similarity 0.88)
- ConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty (NeurIPS 2024 · similarity 0.88)
- Evaluating Numerical Reasoning in Text-to-Image Models (NeurIPS 2024 · similarity 0.88)
- OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation (NeurIPS 2025 · 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.
