Research map / Video / Video segmentation and tracking
Motion-Attentive Transition for Zero-Shot Video Object Segmentation
Tianfei Zhou, Shunzhou Wang, Yi Zhou, Yazhou Yao, Jianwu Li, Ling Shao
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 motion-attentive transitions between motion and appearance features at each encoder stage (ablation) | MATNet · DAVIS-16 | Mean J: 79.5 -> 82.4 | Improved |
| Add motion-attentive transitions between motion and appearance features at each encoder stage (ablation) | MATNet · DAVIS-16 | Mean F: 77.3 -> 80.7 | Improved |
| Use SSA rather than a simple skip layer (ablation) | MATNet · DAVIS-16 | Mean J: 80.7 -> 82.4 | Improved |
| Use SSA rather than a simple skip layer (ablation) | MATNet · DAVIS-16 | Mean F: 79.7 -> 80.7 | Improved |
| Add hard example mining for boundary training (ablation) | MATNet · DAVIS-16 | Mean F: 78.4 -> 80.7 | Improved |
| Add hard example mining for boundary training (ablation) | MATNet · DAVIS-16 | Mean J: 81.4 -> 82.4 | Improved |
| Replace ResNet-101 with ResNet-50 (ablation) | MATNet · DAVIS-16 | Mean J: 82.4 -> 81.1 | Hurt |
| Increase MAT depth from L=0 to L=1 (ablation) | MATNet · DAVIS-16 | Mean J: 79.5 -> 80.6 | Improved |
| Increase MAT depth from L=1 to L=3 (ablation) | MATNet · DAVIS-16 | Mean J: 80.6 -> 81.6 | Improved |
| Increase MAT depth from L=3 to L=5 (ablation) | MATNet · DAVIS-16 | Mean J: 81.6 -> 82.4 | Improved |
| Increase MAT depth from L=5 to L=7 (ablation) | MATNet · DAVIS-16 | Mean J: 82.4 -> 82.2 | Hurt |
| Use the interleaved MATNet encoder rather than independently learned motion and appearance streams | Zero-shot video object segmentation · DAVIS-16 val | Mean J: 80.7 -> 82.4 | Improved |
Closest papers by meaning
- Learning to Segment Moving Objects (IJCV 2018 · similarity 0.89)
- Learning Motion-Appearance Co-Attention for Zero-Shot Video Object Segmentation (ICCV 2021 · similarity 0.89)
- Zero-Shot Video Object Segmentation via Attentive Graph Neural Networks (ICCV 2019 · similarity 0.88)
- The Emergence of Objectness: Learning Zero-shot Segmentation from Videos (NeurIPS 2021 · similarity 0.87)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.86)
- Efficient and Effective Weakly-Supervised Action Segmentation via Action-Transition-Aware Boundary Alignment (CVPR 2024 · similarity 0.85)
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.
