Research map / Vision-language and multimodal / Multimodal LLMs
UniREditBench: A Unified Reasoning-based Image Editing Benchmark
Feng Han, Yibin Wang, Chenglin Li, Zheming Liang, Dianyi Wang, Yang Jiao, Zhipeng Wei, Chao Gong, Cheng Jin, Jiaqi Wang
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 |
|---|---|---|---|
| fine-tune Bagel on UniREdit-Data-100K | Bagel · UniREditBench overall | overall score: 48.01 -> 78.15 | Improved |
| fine-tune Bagel on UniREdit-Data-100K | Bagel-Think · RISEBench | overall accuracy: 9.2% -> 18.3% | Improved |
| fine-tune Bagel on UniREdit-Data-100K | Bagel-Think · KRISBench | overall score: 60.77 -> 65.45 | Improved |
| fine-tune on UniREdit-Data-100K in game-world scenarios versus GPT-4o | GPT-4o · UniREditBench game-world avg | score: 62.07 -> 80.48 | Improved |
| add game-world data to training (ablation) | UniREdit-Bagel · UniREditBench real-world avg | score: 73.81 -> 75.74 | Improved |
| add game-world data to training (ablation) | UniREdit-Bagel · UniREditBench Attribute Modification | score: 74.78 -> 76.73 | Improved |
| add game-world data to training (ablation) | UniREdit-Bagel · UniREditBench Physical Interaction | score: 74.22 -> 76.57 | Improved |
| use weights (0.50,0.30,0.20) instead of equal weights (ablation) | GPT-4.1 evaluator · human-judge Pearson correlation | avg Pearson r: 0.8357 -> 0.8408 | Improved |
| use weights (0.50,0.30,0.20) instead of (0.70,0.30,0.20) (ablation) | GPT-4.1 evaluator · human-judge Pearson correlation | avg Pearson r: 0.8405 -> 0.8408 | No clear effect |
| use dual-reference (text and image) evaluation (ablation) | VLM evaluator · MAE to human scores | MAE | Improved |
| increase training set from 20K to 100K samples (ablation) | UniREdit-Bagel · UniREditBench overall | overall score | Improved |
| enable think mode (reasoning before edit) in Bagel | Bagel · UniREditBench overall | overall score: 48.01 -> 50.96 | Improved |
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.
- Omnigen2 — OmniGen2: Towards Instruction-Aligned Multimodal Generation (CVPR 2026) · 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
- KRIS-Bench: Benchmarking Next-Level Intelligent Image Editing Models (NeurIPS 2025 · similarity 0.91)
- UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched Edits (CVPR 2026 · similarity 0.91)
- Envisioning Beyond the Pixels: Benchmarking Reasoning-Informed Visual Editing (NeurIPS 2025 · similarity 0.90)
- GIR-Bench: Versatile Benchmark for Generating Images with Reasoning (ICLR 2026 · similarity 0.90)
- Learning Action and Reasoning-Centric Image Editing from Videos and Simulation (NeurIPS 2024 · similarity 0.89)
- Factuality Matters: When Image Generation and Editing Meet Structured Visuals (ICLR 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.
