Research map / Image generation / Editing and style transfer
RefEdit: A Benchmark and Method for Improving Instruction-based Image Editing Model on Referring Expressions
Bimsara Pathiraja, Maitreya Patel, Shivam Singh, Yezhou Yang, Chitta Baral
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 20000 RefEdit-Data samples to UltraEdit (RefEdit-SD3) | UltraEdit · RefEdit-Bench Easy | overall VIEScore: 3.4 -> 4.99 | Improved |
| add 20000 RefEdit-Data samples to UltraEdit (RefEdit-SD3) | UltraEdit · RefEdit-Bench Hard | overall VIEScore: 2.59 -> 3.98 | Improved |
| finetune on RefEdit-Data plus MagicBrush data (RefEdit) instead of MagicBrush alone | InstructPix2Pix SD1.5 · RefEdit-Bench Easy | overall VIEScore: 3.67 -> 4.68 | Improved |
| finetune on RefEdit-Data plus MagicBrush data (RefEdit) instead of MagicBrush alone | InstructPix2Pix SD1.5 · RefEdit-Bench Hard | overall VIEScore: 3.56 -> 3.93 | Improved |
| use 20000 targeted synthetic samples rather than 1.2M OmniEdit samples | SD1.5 editing model · PIE-Bench | overall VIEScore: 5.12 -> 5.21 | Improved |
| use larger base model (SDXL CosXLEdit) instead of SD1.5 MagicBrush | image editing baselines · RefEdit-Bench Hard | overall VIEScore: 3.56 -> 2.88 | Hurt |
| use larger base model (SD3 UltraEdit) instead of SD1.5 MagicBrush | image editing baselines · RefEdit-Bench Hard | overall VIEScore: 3.56 -> 2.59 | Hurt |
| use SD3 variant (RefEdit-SD3) instead of SD1.5 RefEdit | RefEdit · PIE-Bench | overall VIEScore: 5.21 -> 5.70 | Improved |
| evaluate on RefEdit-Bench Hard instead of PIE-Bench | RefEdit-SD3 · RefEdit vs PIE benchmarks | overall VIEScore: 5.70 -> 3.98 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- OmniEdit* — OmniEdit: Building Image Editing Generalist Models Through Specialist Supervision (ICLR 2025) · better on 3 of 3 shared results
- InstructPix2Pix — InstructPix2Pix: Learning To Follow Image Editing Instructions (CVPR 2023) · better on 3 of 3 shared results
- HQ-Edit — HQ-Edit: A High-Quality Dataset for Instruction-based Image Editing (ICLR 2025) · better on 2 of 2 shared results
- HIVE — HIVE: Harnessing Human Feedback for Instructional Visual Editing (CVPR 2024) · better on 2 of 2 shared results
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
- ImgEdit: A Unified Image Editing Dataset and Benchmark (NeurIPS 2025 · similarity 0.90)
- Referring Image Editing: Object-level Image Editing via Referring Expressions (CVPR 2024 · similarity 0.90)
- InsightEdit: Towards Better Instruction Following for Image Editing (CVPR 2025 · similarity 0.90)
- AnyEdit: Mastering Unified High-Quality Image Editing for Any Idea (CVPR 2025 · similarity 0.90)
- EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (ICLR 2026 · similarity 0.89)
- Balancing Preservation and Modification: A Region and Semantic Aware Metric for Instruction-Based Image Editing (ICML 2025 · 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.
