Research map / Video / Video segmentation and tracking
Deep Video Matting via Spatio-Temporal Alignment and Aggregation
Yanan Sun, Guanzhi Wang, Qiao Gu, Chi-Keung Tang, Yu-Wing Tai
Check 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 the full ST-FAM video matting framework instead of GCA | Composited test set with full trimaps · Composited test set, full-trimap | SAD: 47.49 -> 40.91 | Improved |
| Add TFA to the basic model (ablation) | Basic video matting model · Composited test set | SAD: 44.26 -> 42.32 | Improved |
| Add TFF to the model with TFA (ablation) | Basic video matting model with TFA · Composited test set | SAD: 42.32 -> 40.91 | Improved |
| Increase neighboring frames from 1 to 2 (ablation) | Temporal aggregation model · Composited test set | SAD: 48.72 -> 46.98 | Improved |
| Increase neighboring frames from 2 to 3 (ablation) | Temporal aggregation model · Composited test set | SAD: 46.98 -> 46.29 | Improved |
| Increase neighboring frames from 3 to 4 (ablation) | Temporal aggregation model · Composited test set | SAD: 46.29 -> 46.30 | No clear effect |
| Replace ST-FAM with naive fusion (ablation) | Video matting model · Composited test set | SAD: 40.91 -> 44.13 | Hurt |
| Replace ST-FAM with cross-attention fusion (ablation) | Video matting model · Composited test set | SAD: 40.91 -> 41.29 | Hurt |
| Replace ST-FAM with flow fusion (ablation) | Video matting model · Composited test set | SAD: 40.91 -> 44.06 | Hurt |
| Provide trimaps every 20 rather than every frame (ablation) | ST-FAM video matting model · Composited test set | SAD: 40.91 -> 43.66 | Hurt |
| Provide trimaps every 40 frames rather than every frame (ablation) | ST-FAM video matting model · Composited test set | SAD: 40.91 -> 52.85 | Hurt |
| Use a single trimap for the entire video rather than frame-by-frame trimaps (ablation) | ST-FAM video matting model · Composited test set | SAD: 40.91 -> 65.33 | Hurt |
Closest papers by meaning
- One-Trimap Video Matting (ECCV 2022 · similarity 0.92)
- End-to-End Video Matting With Trimap Propagation (CVPR 2023 · similarity 0.91)
- Deep Image Matting With Sparse User Interactions (TPAMI 2023 · similarity 0.91)
- MatAnyone: Stable Video Matting with Consistent Memory Propagation (CVPR 2025 · similarity 0.89)
- Training Matting Models Without Alpha Labels (AAAI 2025 · similarity 0.89)
- Revisiting Context Aggregation for Image Matting (ICML 2024 · similarity 0.88)
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.
