Research map / Image generation / Editing and style transfer
I2I-Bench: A Comprehensive Benchmark Suite for Image-to-Image Editing Models
Juntong Wang, Jiarui Wang, Huiyu Duan, Jiaxiang Kang, Guangtao Zhai, Xiongkuo Min
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 |
|---|---|---|---|
| use specialist model (ArtiMuse) instead of pure LMM for aesthetic quality (ablation) | I2I-Bench evaluation pipeline · SE Aesthetic-Quality | Pearson rho with human preference: -0.0643 -> 0.9889 | Improved |
| use hybrid LMM-specialist feature matching instead of pure LMM for subject identity (ablation) | I2I-Bench evaluation pipeline · SE Subject-Identity | Pearson rho with human preference: -0.3494 -> 0.9133 | Improved |
| use specialist Q-Insight instead of pure LMM for image quality in multi-image editing (ablation) | I2I-Bench evaluation pipeline · ME Image-Quality | Pearson rho with human preference: -0.9202 -> 0.6579 | Improved |
| use hybrid pipeline instead of pure LMM overall (ablation) | I2I-Bench evaluation pipeline · SE overall | Pearson rho with human preference: 0.7277 -> 0.9425 | Improved |
| use hybrid pipeline instead of pure LMM overall (ablation) | I2I-Bench evaluation pipeline · ME overall | Pearson rho with human preference: 0.6829 -> 0.8688 | Improved |
| use LMM VQA-5level versus specialist for Blending-Naturalness and Instruction-Following (same method used) (ablation) | I2I-Bench evaluation pipeline · SE Blending-Naturalness | Pearson rho with human preference: 0.8866 -> 0.8866 | No clear effect |
| I2I-Bench hybrid evaluation instead of LMM4Edit | evaluation metric · SE 15 dimensions average | Pearson rho with human preference: 0.5968 -> 0.9425 | Improved |
| LMM4Edit versus I2I-Bench on Spatial-Accuracy | evaluation metric · SE Spatial-Accuracy | Pearson rho with human preference: 0.9655 -> 0.9303 | Hurt |
| switch from single-image to multi-image editing | Qwen-Image-Edit-2509 · I2I-Bench Blending-Naturalness | score: 0.807 -> 0.556 | Hurt |
| switch from single-image to multi-image editing | Qwen-Image-Edit-2509 · I2I-Bench Instruction-Following | score: 0.930 -> 0.744 | Hurt |
| increase task cognitive complexity from Local Editing to World Knowledge | Qwen-Image-Edit-2509 · I2I-Bench SE Instruction-Following | score: 0.992 -> 0.805 | Hurt |
Closest papers by meaning
- I2EBench: A Comprehensive Benchmark for Instruction-based Image Editing (NeurIPS 2024 · similarity 0.93)
- ImgEdit: A Unified Image Editing Dataset and Benchmark (NeurIPS 2025 · similarity 0.92)
- Towards Scalable Human-aligned Benchmark for Text-guided Image Editing (CVPR 2025 · similarity 0.89)
- KRIS-Bench: Benchmarking Next-Level Intelligent Image Editing Models (NeurIPS 2025 · similarity 0.89)
- EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (ICLR 2026 · similarity 0.88)
- Balancing Preservation and Modification: A Region and Semantic Aware Metric for Instruction-Based Image Editing (ICML 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.
