Research map / Image generation / Text-to-image diffusion
SDMatte: Grafting Diffusion Models for Interactive Matting
Longfei Huang, Yu Liang, Hao Zhang, Jinwei Chen, Wei Dong, Lunde Chen, Wanyu Liu, Bo Li, Peng-Tao Jiang
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 visual prompt cross-attention in middle block only (ablation) | SDMatte · AIM-500 and RefMatte-RW-100 (point, box) | average improvement: +11.67% | Improved |
| add visual prompt cross-attention in down blocks only (ablation) | SDMatte · AIM-500 and RefMatte-RW-100 | average improvement: -4.06% | Hurt |
| add visual prompt cross-attention in up blocks only (ablation) | SDMatte · AIM-500 and RefMatte-RW-100 | average improvement: -7.77% | Hurt |
| add visual prompt cross-attention in all blocks (ablation) | SDMatte · AIM-500 and RefMatte-RW-100 | average improvement: -0.65% | No clear effect |
| add opacity embedding (ablation) | SDMatte · AIM-500 (point) MSE | MSE: 0.0169 -> 0.0149 | Improved |
| add coordinate embedding (ablation) | SDMatte · RefMatte-RW-100 (point) MSE | MSE: 0.0115 -> 0.0104 | Improved |
| add both opacity and coordinate embeddings (ablation) | SDMatte · AIM-500 and RefMatte-RW-100 | average improvement: +10.20% | Improved |
| add masked self-attention in all U-Net blocks (ablation) | SDMatte · AIM-500 and RefMatte-RW-100 | average improvement: +44.43% | Improved |
| add masked self-attention in up blocks only (ablation) | SDMatte · AIM-500 and RefMatte-RW-100 | average improvement: +38.10% | Improved |
| add masked self-attention in down blocks only (ablation) | SDMatte · AIM-500 and RefMatte-RW-100 | average improvement: -5.12% | Hurt |
| replace VAE and U-Net with TinyVAE and BK-U-Net (LiteSDMatte) | SDMatte · AIM-500 (box prompt) | MSE: 0.0049 -> 0.0056 | Hurt |
| train on set 2 with box prompt (SDMatte*) instead of set 1 | SDMatte · AIM-500 (box prompt) | MSE: 0.0049 -> 0.0036 | Improved |
Later papers that compared with this method
1 later accepted paper reports SDMatte as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Edit2Perceive: Image Editing Diffusion Models Are Strong Dense Perceivers (CVPR 2026 · better on 2 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SDMatte++: Grafting High-Resolution Diffusion Models for Interactive Matting (TPAMI 2026 · similarity 0.97)
- Unifying Automatic and Interactive Matting with Pretrained ViTs (CVPR 2024 · similarity 0.90)
- In-Context Matting (CVPR 2024 · similarity 0.89)
- Revisiting Context Aggregation for Image Matting (ICML 2024 · similarity 0.88)
- DRIP: Unleashing Diffusion Priors for Joint Foreground and Alpha Prediction in Image Matting (NeurIPS 2024 · similarity 0.87)
- Training Matting Models Without Alpha Labels (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.
