Research map / Video / Video segmentation and tracking
End-to-End Video Matting With Trimap Propagation
Wei-Lun Huang, Ming-Sui Lee
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 |
|---|---|---|---|
| Use FTP-VM rather than the strongest reported baseline | video matting with trimap propagation · VM108 | MSE: 2.92 -> 5.19 | Hurt |
| Use FTP-VM rather than the strongest reported baseline | video matting with trimap propagation · VM240k | MSE: 0.56 -> 0.45 | Improved |
| Use FTP-VM rather than the strongest reported baseline | video matting with trimap propagation · Real Human | MSE: 2.38 -> 2.80 | Hurt |
| Combine trimap propagation and matting in one network rather than use a two-stage network (ablation) | FTP-VM · VM108 | MSE: 8.80 -> 10.21 | Hurt |
| Use separate, specially designed segmentation and matting decoders instead of the same decoders (ablation) | FTP-VM · VM108 | MSE: 9.72 -> 5.19 | Improved |
| Replace an additional backbone with vanilla convolutions in trimap fusion (ablation) | FTP-VM · VM108 | MSE: 11.34 -> 8.12 | Improved |
| Replace vanilla convolutions with gated convolutions in trimap fusion (ablation) | FTP-VM · VM108 | MSE: 8.12 -> 5.19 | Improved |
| Add segmentation consistency loss (ablation) | FTP-VM trimap segmentation · VM108 | MSE: 11.66 -> 5.19 | Improved |
| Apply consistency loss to all classes rather than only the correct class (ablation) | FTP-VM trimap segmentation · VM108 | MSE: 6.19 -> 8.75 | Hurt |
| Weight consistency loss differently for correct and incorrect classes instead of applying it unweighted to all classes (ablation) | FTP-VM trimap segmentation · VM108 | MSE: 8.75 -> 5.19 | Improved |
| Use weighted consistency loss across all classes rather than correct-class-only consistency (ablation) | FTP-VM trimap segmentation · VM108 | dtSSD: 2.28 -> 2.46 | Hurt |
| Use the end-to-end FTP-VM instead of the two-model approaches | trimap propagation and video matting · inference | FPS: 5.11 -> 40.16 | Improved |
Closest papers by meaning
- One-Trimap Video Matting (ECCV 2022 · similarity 0.94)
- Deep Video Matting via Spatio-Temporal Alignment and Aggregation (CVPR 2021 · similarity 0.91)
- Deep Image Matting With Sparse User Interactions (TPAMI 2023 · similarity 0.90)
- Matting Anything 2: Towards Video Matting for Anything (ICLR 2026 · similarity 0.89)
- MatAnyone: Stable Video Matting with Consistent Memory Propagation (CVPR 2025 · similarity 0.88)
- Training Matting Models Without Alpha Labels (AAAI 2025 · 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.
