Research map / Detection and segmentation / Open-set and few-shot detection
DreamMix: Decoupling Object Attributes for Enhanced Editability in Customized Image Inpainting
Yicheng Yang, Pengxiang Li, Lu Zhang, Liqian Ma, Ping Hu, Siyu Du, Yunzhi Zhuge, Jia Hao Xu, Huchuan Lu
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 Disentangled Inpainting Framework (DIF) (ablation) | DreamBooth baseline on SDXL inpainting · Attribute editing | CLIP-T: 0.235 -> 0.265 | Improved |
| add DIF (ablation) | DreamBooth baseline on SDXL inpainting · Identity preservation | CLIP-I: 0.644 -> 0.723 | Improved |
| add Textual Attribute Substitution (TAS) on top of DIF (ablation) | DreamMix SDXL · Attribute editing | CLIP-T: 0.265 -> 0.275 | Improved |
| add TAS on top of DIF (ablation) | DreamMix SDXL · Attribute editing | CLIP-I: 0.696 -> 0.685 | Hurt |
| add Attribute Decoupling Mechanism (ADM) on top of DIF and TAS (ablation) | DreamMix SDXL · Attribute editing | CLIP-T: 0.275 -> 0.289 | Improved |
| add TAS and ADM on top of DIF (ablation) | DreamMix SDXL · Identity preservation | CLIP-I: 0.723 -> 0.728 | No clear effect |
| use ChatGPT-4o versus smaller open VLMs in ADM and TAS (ablation) | DreamMix SDXL · Attribute editing | CLIP-T: 0.287, 0.290, 0.287 vs 0.289 | No clear effect |
| increase regularization images from 10 to 50 (ablation) | DreamMix SDXL · Attribute editing | CLIP-T: 0.285 -> 0.290 | No clear effect |
| remove loss reweighting (tau1=tau2=1.0 instead of 1.5/0.7) (ablation) | DreamMix SDXL · Attribute editing | CLIP-T: 0.289 -> 0.286 | No clear effect |
| add DreamMix components when training on a single image (ablation) | DreamBooth* baseline · Attribute editing | CLIP-T: 0.235 -> 0.284 | Improved |
| use only GCH (lambda=1) or only LCG (lambda=0) instead of lambda=0.7 (ablation) | DreamMix DIF · small object inpainting | qualitative | Hurt |
| use DreamMix instead of DreamBooth on same SDXL inpainting backbone | SDXL · Attribute editing | CLIP-T: 0.235 -> 0.289 | Improved |
Closest papers by meaning
- Paint by Inpaint: Learning to Add Image Objects by Removing Them First (CVPR 2025 · similarity 0.87)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.85)
- DesignEdit: Unify Spatial-Aware Image Editing via Training-free Inpainting with a Multi-Layered Latent Diffusion Framework (AAAI 2025 · similarity 0.85)
- Object-Aware Inversion and Reassembly for Image Editing (ICLR 2024 · similarity 0.85)
- An Item Is Worth a Prompt: Versatile Image Editing with Disentangled Control (AAAI 2025 · similarity 0.85)
- PAIR Diffusion: A Comprehensive Multimodal Object-Level Image Editor (CVPR 2024 · similarity 0.84)
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.
