Research map / Image generation / Editing and style transfer
Clothe and Pose
Nakul Sharma, Aayush Bansal, Minh Vo
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 |
|---|---|---|---|
| joint single-model clothing and posing instead of try-on then reposing pipeline | SDXL-based multi-stream model · Clothe and Pose, Front->Back | LPIPS: 0.277 -> 0.166 vs CatVTON+Leffa | Improved |
| provide front and back garment images at inference for reposing (ablation) | Ours (reposing) · Reposing, Front->Back | PSNR: 19.972 -> 20.801 | Improved |
| provide front and back garment images at inference for reposing (ablation) | Ours (reposing) · Reposing, Front->Front | PSNR: 21.207 -> 22.439 | Improved |
| remove text cross-attention from SDXL UNet and condition on pose and garment images only | SDXL UNet · Clothe and Pose | parameters: removes roughly 0.5 B parameters | Improved |
| multi-stream blocks with joint attention instead of spatial concatenation of conditions | SDXL UNet · Clothe and Pose | network capacity use | Improved |
| add stream embedding and position embeddings to condition tokens | Joint attention · early experiments | learning speed | Improved |
| randomly drop garment patches in user image in stage-1 training | Stage-1 pretraining · Clothe and Pose | reliance on garment conditions | Improved |
| use Qwen-Image-Edit (20B) instead of the 5B model for joint clothing and posing | baseline · Clothe and Pose, Front->Back | LPIPS: 0.166 -> 0.340 | Hurt |
| text-guided Flux.1-Kontext reposing | pipeline baseline · Reposing, Front->Back | LPIPS: 0.287 vs ours 0.125 | Hurt |
| train reposing models on DeepFashion benchmark (Leffa, CFLD) | Leffa/CFLD · Reposing | identity preservation | Hurt |
Closest papers by meaning
- FashionTailor: Controllable Clothing Editing for Human Images with Appearance Preserving (AAAI 2025 · similarity 0.87)
- DressWild: Feed-Forward Pose-Agnostic Garment Sewing Pattern Generation from In-the-Wild Images (NeurIPS 2026 · similarity 0.87)
- Controllable Human Image Generation with Personalized Multi-Garments (CVPR 2025 · similarity 0.86)
- FreeCloth: Free-form Generation Enhances Challenging Clothed Human Modeling (CVPR 2025 · similarity 0.86)
- UniHuman: A Unified Model For Editing Human Images in the Wild (CVPR 2024 · similarity 0.86)
- DreamFit: Garment-Centric Human Generation via a Lightweight Anything-Dressing Encoder (AAAI 2025 · similarity 0.85)
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.
