Research map / Vision-language and multimodal / Multimodal LLMs
GRADE: Benchmarking Discipline-Informed Reasoning in Image Editing
Mingxin Liu, Ziqian Fan, Zhaokai Wang, Leyao Gu, Zirun Zhu, YiguoHe YiguoHe, Yuchen Yang, Changyao Tian, Xiangyu Zhao, Ning Liao, Shaofeng Zhang, Qibing Ren and 4 more
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 |
|---|---|---|---|
| make instructions explicit (ablation) | Nano Banana 2 · GRADE 68-sample subset | accuracy (%): 35.3 -> 65.7 | Improved |
| make instructions explicit (ablation) | Nano Banana 2 · GRADE 68-sample subset | Discipline Reasoning: 67.9 -> 89.7 | Improved |
| make instructions explicit (ablation) | Qwen-Edit-2511 · GRADE 68-sample subset | Discipline Reasoning: 18.9 -> 44.7 | Improved |
| make instructions explicit (ablation) | Qwen-Edit-2511 · GRADE 68-sample subset | accuracy (%): 1.5 -> 8.8 | Improved |
| make instructions explicit (ablation) | Nano Banana 2 · GRADE 68-sample subset | Readability: 95.6 -> 95.5 | No clear effect |
| use Gemini-3-Flash as judge instead of GPT-5 (ablation) | evaluation pipeline · Human alignment (68 samples) | Reasoning MAE: 0.1519 -> 0.1194 | Improved |
| use Gemini-3-Flash as judge instead of Qwen3-VL-235B (ablation) | evaluation pipeline · Human alignment (68 samples) | Reasoning MAE: 0.1798 -> 0.1194 | Improved |
| add thinking to Bagel | Bagel · GRADE (520 samples) | accuracy (%): 0.6 -> 0.2 | Hurt |
| add think+reflect to Step-1X | Step-1X · GRADE (520 samples) | accuracy (%): 1.0 -> 2.3 | Improved |
| add think to Step-1X | Step-1X · GRADE (520 samples) | accuracy (%): 1.0 -> 1.4 | Improved |
Closest papers by meaning
- KRIS-Bench: Benchmarking Next-Level Intelligent Image Editing Models (NeurIPS 2025 · similarity 0.92)
- Envisioning Beyond the Pixels: Benchmarking Reasoning-Informed Visual Editing (NeurIPS 2025 · similarity 0.91)
- GIR-Bench: Versatile Benchmark for Generating Images with Reasoning (ICLR 2026 · similarity 0.90)
- UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched Edits (CVPR 2026 · similarity 0.90)
- Learning Action and Reasoning-Centric Image Editing from Videos and Simulation (NeurIPS 2024 · similarity 0.90)
- Reasoning to Edit: Hypothetical Instruction-Based Image Editing with Visual Reasoning (ICML 2026 · similarity 0.89)
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.
