Research map / Video / Video generation
VideoMaMa: Mask-Guided Video Matting via Generative Prior
Sangbeom Lim, Seoung Wug Oh, Jiahui Huang, Heeji Yoon, Seungryong Kim, Joon-Young 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 |
|---|---|---|---|
| add MA-V pseudo-labeled data to existing matting datasets for SAM2 fine-tuning (ablation) | SAM2-Matte · V-HIM60 Hard | MAD: 7.5798 -> 2.6112 | Improved |
| train on MA-V alone instead of existing datasets alone (ablation) | SAM2-Matte · V-HIM60 Hard | MAD: 7.5798 -> 3.1830 | Improved |
| add existing synthetic matting datasets to MA-V (ablation) | SAM2-Matte · DAVIS val | J&F: 87.9 -> 85.9 | Hurt |
| add existing synthetic matting datasets to MA-V (ablation) | SAM2-Matte · V-HIM60 Hard | MAD: 3.1830 -> 2.6112 | Improved |
| add stage 2 temporal training to stage 1 spatial training (ablation) | VideoMaMa · YouTubeMatte, downsample 32x | MAD: 3.7620 -> 1.2552 | Improved |
| add stage 1 high-resolution spatial training to video-only stage 2 (with DINO) (ablation) | VideoMaMa · YouTubeMatte, downsample 32x | MAD: 1.2350 -> 1.2552 (without DINO) | Mixed |
| add DINOv3 semantic knowledge injection (ablation) | VideoMaMa · YouTubeMatte, downsample 32x | MAD: 1.2552 -> 1.0292 | Improved |
| add DINOv3 semantic knowledge injection (ablation) | VideoMaMa · YouTubeMatte, hard polygon masks | MAD: 1.9975 -> 1.4042 | Improved |
| add DINOv3 semantic knowledge injection (ablation) | VideoMaMa · YouTubeMatte, SAM2 masks | MAD: 1.9404 -> 1.7370 | Improved |
| vary number of frames per inference from 1 to 24 (ablation) | VideoMaMa · YouTubeMatte, downsample 32x | MAD (refined): 1.1024 (1 frame) -> 1.0237 (24 frames) | No clear effect |
| use VideoMaMa to refine SAM2 masks versus raw SAM2 masks (first-frame setting) | SAM2+VideoMaMa · V-HIM60 Hard | MAD: 7.8517 -> 2.7757 | Improved |
| use SAM2-Matte versus MatAnyone | SAM2-Matte · V-HIM60 Hard | MAD: 5.7195 -> 2.6112 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MatAnyone — MatAnyone: Stable Video Matting with Consistent Memory Propagation (CVPR 2025) · better on 5 of 5 shared results
- SAM2 — SAM 2: Segment Anything in Images and Videos (ICLR 2025) · better on 3 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
- VideoMaMa++: Temporally Consistent Video Matting via Preserve-and-Refine Tokens (NeurIPS 2026 · similarity 0.94)
- Static-to-Dynamic: Animating Still Mattes via Generative Motion for Video Matting (NeurIPS 2026 · similarity 0.93)
- MaGGIe: Masked Guided Gradual Human Instance Matting (CVPR 2024 · similarity 0.90)
- MatAnyone: Stable Video Matting with Consistent Memory Propagation (CVPR 2025 · similarity 0.90)
- Training Matting Models Without Alpha Labels (AAAI 2025 · similarity 0.89)
- In-Context Matting (CVPR 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.
