Research map / Architectures and efficiency / Attention and transformers
MatteFormer: Transformer-Based Image Matting via Prior-Tokens
GyuTae Park, SungJoon Son, JaeYoung Yoo, SeHo Kim, Nojun Kwak
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 |
|---|---|---|---|
| Use MatteFormer instead of SIM | Composition-1k matting · Composition-1k test | SAD: 28.0 -> 23.8 | Improved |
| Use MatteFormer instead of MG Matting-trimap,res50* | Composition-1k matting · Composition-1k test | MSE ($10^{-3}$): 5.4 -> 4.0 | Improved |
| Use MatteFormer instead of SIM | Composition-1k matting · Composition-1k test | Grad: 10.8 -> 8.7 | Improved |
| Use MatteFormer instead of MG Matting-trimap,res50* | Composition-1k matting · Composition-1k test | Conn: 24.3 -> 18.9 | Improved |
| Use MatteFormer instead of MG Matting-trimap* | Distinctions-646 matting · Distinctions-646 test | SAD: 23.9 -> 21.9 | Improved |
| Use MatteFormer instead of MG Matting-trimap* | Distinctions-646 matting · Distinctions-646 test | MSE ($10^{-3}$): 7.4 -> 6.6 | Improved |
| Use MatteFormer instead of MG Matting-trimap* | Distinctions-646 matting · Distinctions-646 test | Grad: 14.0 -> 11.2 | Improved |
| Use MatteFormer instead of MG Matting-trimap* | Distinctions-646 matting · Distinctions-646 test | Conn: 22.4 -> 20.5 | Improved |
| Add a GAP prior-token (ablation) | Swin Transformer Tiny matting baseline · Composition-1k | SAD: 26.43 -> 25.30 | Improved |
| Replace the GAP token with an unknown-region prior-token (ablation) | Swin Transformer Tiny matting baseline · Composition-1k | SAD: 25.30 -> 24.70 | Improved |
| Add foreground and background prior-tokens alongside the unknown-region prior-token (ablation) | Swin Transformer Tiny matting baseline · Composition-1k | SAD: 24.70 -> 24.19 | Improved |
| Add prior-memory to the three trimap-region prior-tokens (ablation) | Swin Transformer Tiny matting baseline · Composition-1k | SAD: 24.19 -> 23.80 | Improved |
Later papers that compared with this method
1 later accepted paper reports MatteFormer as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Revisiting Context Aggregation for Image Matting (ICML 2024 · better on 2 of 2 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
- Memory Efficient Matting with Adaptive Token Routing (AAAI 2025 · similarity 0.89)
- Privileged Prior Information Distillation for Image Matting (AAAI 2024 · similarity 0.88)
- Revisiting Context Aggregation for Image Matting (ICML 2024 · similarity 0.87)
- MaGGIe: Masked Guided Gradual Human Instance Matting (CVPR 2024 · similarity 0.87)
- Training Matting Models Without Alpha Labels (AAAI 2025 · similarity 0.87)
- Unifying Automatic and Interactive Matting with Pretrained ViTs (CVPR 2024 · 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.
