Research map / Image generation / Autoregressive and latent generation
Referring Image Matting
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 CLIPMat rather than the adapted referring segmentation baselines | Referring image matting · RefMatte keyword setting | SAD: 17.75 -> 9.91 | Helped |
| Add the matting refiner (ablation) | CLIPMat · RefMatte keyword setting | SAD: 9.91 -> 9.13 | Helped |
| Use the larger ViT-L/14 backbone instead of ViT-B/16 | CLIPMat · RefMatte keyword setting | SAD: 9.91 -> 8.51 | Helped |
| Fine-tune the pretrained CLIP encoders instead of freezing them (ablation) | CLIPMat · RefMatte keyword test | SAD: 20.05 -> 9.91 | Helped |
| Pretrain on VGPhraseCut rather than train directly on RefMatte (ablation) | CLIPMat · RefMatte keyword test | SAD: 9.91 -> 11.33 | Hurt |
| Use matting-related prompt templates, including “a photograph of a {keyword}” (ablation) | CLIPMat · RefMatte keyword test | SAD: 9.91 -> 9.82 | Helped |
| Use RPE-1 rather than a basic expression (ablation) | CLIPMat · RefMatte expression setting | SAD: 59.20 -> 25.98 | Helped |
| Use free-form human expressions rather than structured position descriptions (ablation) | CLIPMat · RefMatte-RW100 | SAD: 85.08 -> 157.34 | Hurt |
Closest papers by meaning
- In-Context Matting (CVPR 2024 · similarity 0.91)
- Revisiting Context Aggregation for Image Matting (ICML 2024 · similarity 0.88)
- Unifying Automatic and Interactive Matting with Pretrained ViTs (CVPR 2024 · similarity 0.88)
- Training Matting Models Without Alpha Labels (AAAI 2025 · similarity 0.88)
- Privileged Prior Information Distillation for Image Matting (AAAI 2024 · similarity 0.86)
- MAGICK: A Large-scale Captioned Dataset from Matting Generated Images using Chroma Keying (CVPR 2024 · similarity 0.86)
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.