Research map / Video / Video segmentation and tracking
MOVE: Motion-Guided Few-Shot Video Object Segmentation
Kaining Ying, Hengrui Hu, Henghui Ding
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 |
|---|---|---|---|
| replace mask pooling with frame-differencing motion extractor (ablation) | DMA · MOVE OS 2-way-1-shot (ResNet50) | J&F: 41.3 -> 46.8 | Improved |
| replace mask adapter with frame-differencing motion extractor (ablation) | DMA · MOVE OS 2-way-1-shot (ResNet50) | J&F: 43.4 -> 46.8 | Improved |
| add motion prototype to appearance prototype (ablation) | DMA · MOVE OS 2-way-1-shot (ResNet50) | J&F: 36.5 -> 46.8 | Improved |
| add appearance prototype to motion prototype (ablation) | DMA · MOVE OS 2-way-1-shot (ResNet50) | J&F: 43.8 -> 46.8 | Improved |
| add auxiliary classification on object and motion prototypes (ablation) | DMA · MOVE OS 2-way-1-shot (ResNet50) | J&F: 43.8 -> 46.8 | Improved |
| add only object classification (no motion classification) (ablation) | DMA · MOVE OS 2-way-1-shot (ResNet50) | J&F: 43.8 -> 43.5 | No clear effect |
| add only motion classification (no object classification) (ablation) | DMA · MOVE OS 2-way-1-shot (ResNet50) | J&F: 43.8 -> 44.2 | No clear effect |
| use VideoSwin-T instead of ResNet50 backbone (ablation) | DMA · MOVE OS 5-way-1-shot | J&F: 40.2 -> 41.4 | Improved |
| supply temporal modeling (self-attention across frames) to HPAN prototypes | HPAN · MOVE | J&F: 44.4 -> 46.3 | Improved |
| evaluate image FSS method SCCAN on MOVE instead of YouTube-VIS | SCCAN · MOVE vs YouTube-VIS | J&F: 62.3 -> 40.6 | Hurt |
| use non-overlapping instead of overlapping split | DMA · MOVE 2-way-1-shot (ResNet50) | J&F: 50.1 -> 46.0 | Hurt |
| provide oracle ground-truth mask | DMA · MOVE OS 2-way-1-shot (ResNet50) | J&F: 46.8 -> 74.3 | Improved |
Closest papers by meaning
- Learning Motion Patterns in Videos (CVPR 2017 · similarity 0.91)
- Learning to Segment Moving Objects (IJCV 2018 · similarity 0.90)
- Segment Any Motion in Videos (CVPR 2025 · similarity 0.90)
- Instance-Level Moving Object Segmentation from a Single Image with Events (IJCV 2025 · similarity 0.89)
- GMOS: Grounding Moving Object Segmentation in 3D Space and Time (NeurIPS 2026 · similarity 0.89)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.89)
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.
