Research map / Image generation / Editing and style transfer
Object Fusion via Diffusion Time-step for Customized Image Editing with Single Example
Xue Song, Zhongqi Yue, Jiequan Cui, Hanwang Zhang, Jingjing Chen
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 learned time-step and patch coefficients to LoRA fusion (ablation) | LoRA fusion · Customized image editing (50 pairs) | CLIP-I / DINO: CLIP-I 0.797 -> 0.828; DINO 0.788 -> 0.806 | Improved |
| add learned time-step and patch coefficients to LoRA fusion (ablation) | LoRA fusion · Customized image editing (50 pairs) | CLIP-T: 0.321 -> 0.316 | Hurt |
| add patch-specific coefficients to time-step-only fusion (ablation) | TimeFusion · Customized image editing (50 pairs) | DINO: 0.795 -> 0.806 | Improved |
| add time-step-specific coefficients to patch-only fusion (ablation) | TimeFusion · Customized image editing (50 pairs) | CLIP-T / DINO: CLIP-T 0.297 -> 0.316; DINO 0.768 -> 0.806; CLIP-I 0.871 -> 0.828 | Mixed |
| increase patch coefficient size from 1x1 to 8x8 (ablation) | TimeFusion · Customized image editing (qualitative) | fidelity (visual) | Improved |
| increase number of time-step splits K from 5 to 20 (ablation) | TimeFusion · Customized image editing (qualitative) | fidelity (visual) | Improved |
| increase K from 20 to 50 (ablation) | TimeFusion · Customized image editing (qualitative) | fidelity (visual) | Hurt |
| use TimeFusion instead of AnyDoor | customized image editing · User study (2,550 answers) | preference rate: 10.2% -> 89.6% | Improved |
Closest papers by meaning
- PAIR Diffusion: A Comprehensive Multimodal Object-Level Image Editor (CVPR 2024 · similarity 0.89)
- LoRACLR: Contrastive Adaptation for Customization of Diffusion Models (CVPR 2025 · similarity 0.88)
- Orthogonal Adaptation for Modular Customization of Diffusion Models (CVPR 2024 · similarity 0.88)
- Energy-Guided Optimization for Personalized Image Editing with Pretrained Text-to-Image Diffusion Models (AAAI 2025 · similarity 0.87)
- LoRAShop: Training-Free Multi-Concept Image Generation and Editing with Rectified Flow Transformers (NeurIPS 2025 · similarity 0.87)
- PixelMan: Consistent Object Editing with Diffusion Models via Pixel Manipulation and Generation (AAAI 2025 · 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.
