Research map / Image generation / Text-to-image diffusion
EditHF-1M: A Million-Scale Rich Human Preference Feedback for Image Editing
Zitong Xu, Huiyu Duan, Zhongpeng Ji, Xinyun Zhang, Yutao Liu, Xiongkuo Min, Ke Gu, Jian Zhang, Shusong Xu, Jinwei Chen, Bo Li, Guangtao Zhai
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 MLP score head (ablation) | EditHF (InternVL3.5, pointwise) · EditHF-1M Visual Quality | SRCC (global): 0.7184 -> 0.7762 | Improved |
| add pairwise training on top of pointwise training (ablation) | EditHF · EditHF-1M Visual Quality | SRCC (global): 0.7762 -> 0.8548 | Improved |
| add pointwise training on top of pairwise training (ablation) | EditHF · EditHF-1M Visual Quality | SRCC (global): 0.8018 -> 0.8548 | Improved |
| pairwise-only vs pointwise-only training (ablation) | EditHF · EditHF-1M Visual Quality | pairwise accuracy: 0.7826 (pointwise) vs 0.8663 (pairwise) | Mixed |
| use InternVL3.5 instead of InternVL3 backbone (ablation) | EditHF · EditHF-1M Visual Quality | SRCC (global): 0.8316 -> 0.8548 | Improved |
| use InternVL3.5 instead of Qwen3-VL backbone (ablation) | EditHF · EditHF-1M Editing Alignment | SRCC (global): 0.8231 -> 0.8488 | Improved |
| fine-tune MLLM on EditHF-1M | InternVL3 (8B) · EditHF-1M Visual Quality | SRCC (global): 0.2462 -> 0.8044 | Improved |
| use EditHF vs EditReward on unseen benchmark | EditHF · ImagenHub | SRCC: 0.3520 -> 0.4526 | Improved |
| fine-tune Qwen-Image-Edit with Self-DPO from EditHF | Qwen-Image-Edit · EditHF-1M human overall | overall score: 75.39 -> 78.36 | Improved |
| fine-tune Qwen-Image-Edit with Global-DPO using multi-model pairs | Qwen-Image-Edit · EditHF-1M human overall | overall score: 75.39 -> 84.26 | Improved |
| Global-DPO instead of Self-DPO (ablation) | Qwen-Image-Edit · EBench-18K human overall | overall score: 73.80 -> 74.18 | Improved |
| Self-DPO on AROURA-Bench judged by EditHF | Qwen-Image-Edit · AROURA-Bench EditHF score | overall score: 68.47 -> 69.26 | Improved |
Closest papers by meaning
- HIVE: Harnessing Human Feedback for Instructional Visual Editing (CVPR 2024 · similarity 0.90)
- Towards Scalable Human-aligned Benchmark for Text-guided Image Editing (CVPR 2025 · similarity 0.88)
- ImgEdit: A Unified Image Editing Dataset and Benchmark (NeurIPS 2025 · similarity 0.88)
- Multi-Reward as Condition for Instruction-based Image Editing (ICLR 2025 · similarity 0.87)
- I2EBench: A Comprehensive Benchmark for Instruction-based Image Editing (NeurIPS 2024 · similarity 0.87)
- Multimodal Large Language Models Make Text-to-Image Generative Models Align Better (NeurIPS 2024 · similarity 0.87)
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.
