Research map / Image generation / Editing and style transfer
FineEdit: Fine-Grained Image Edit with Bounding Box Guidance
Haohang Xu, Lin Liu, Zhibo Zhang, Rong Cong, Xiaopeng Zhang, Qi Tian
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 early fusion only (bbox masked image concatenated at input) instead of early plus deep fusion (ablation) | FineEdit · FineEdit-1k Bench | background PSNR: 34.4 -> 15.9 | Hurt |
| add deep fusion side adapter on top of early fusion (ablation) | FineEdit · FineEdit-1k Bench | background PSNR: 15.9 -> 34.4 | Improved |
| add early fusion on top of deep fusion (ablation) | FineEdit · FineEdit-1k Bench | Prompt Compliance: 4.52 -> 4.67 | Improved |
| apply RL post-training with global reward (ablation) | FineEdit · FineEdit-1k Bench | background PSNR: 34.4 -> 32.7 | Hurt |
| apply RL post-training with global reward (ablation) | FineEdit · FineEdit-1k Bench | Prompt Compliance: 4.67 -> 4.74 | Improved |
| use decoupled ROI/background reward instead of global reward in RL (ablation) | FineEdit · FineEdit-1k Bench | background PSNR: 32.7 -> 34.7 | Improved |
| use decoupled reward instead of global reward in RL (ablation) | FineEdit · FineEdit-1k Bench | Prompt Compliance: 4.74 -> 4.80 | Improved |
| apply bbox-guided FineEdit vs Qwen-Image-Edit-2509 (baseline for FineEdit) on prompt compliance | Qwen-Image-Edit · FineEdit-1k Bench | Prompt Compliance: 4.56 -> 4.67 | Improved |
| FineEdit versus Flux-fill-dev and Asuka-flux on Out-of-Box Retention | FineEdit · FineEdit-1k Bench | Out-of-Box Retention: 5.00 -> 4.80 | Hurt |
| FineEdit pre-training only versus Qwen-Image-Edit-2509 on ImgEdit-Bench overall | Qwen-Image-Edit-2509 · ImgEdit-Bench (single-turn) | Overall: 4.35 -> 4.50 | Improved |
| FineEdit versus Qwen-Image-Edit-2509 on GEdit-Bench-EN | Qwen-Image-Edit-2509 · GEdit-Bench-EN | G_O: 7.54 -> 7.97 (G_O); G_PQ 7.86 -> 7.47 | Mixed |
Closest papers by meaning
- FireEdit: Fine-grained Instruction-based Image Editing via Region-aware Vision Language Model (CVPR 2025 · similarity 0.90)
- ImgEdit: A Unified Image Editing Dataset and Benchmark (NeurIPS 2025 · similarity 0.89)
- UltraEdit: Instruction-based Fine-Grained Image Editing at Scale (NeurIPS 2024 · similarity 0.88)
- An Item Is Worth a Prompt: Versatile Image Editing with Disentangled Control (AAAI 2025 · similarity 0.88)
- PIXELS: Progressive Image Xemplar-based Editing with Latent Surgery (AAAI 2025 · similarity 0.88)
- Towards Efficient Diffusion-Based Image Editing with Instant Attention Masks (AAAI 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.
