Research map / Video / Video segmentation and tracking
Improving Video Segmentation via Dynamic Anchor Queries
Yikang Zhou, Tao Zhang, Xiangtai Li, Shunping Ji, Shuicheng Yan
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 emergence DAQ without emergence simulation (ablation) | DVIS baseline · OVIS (R50, 40K) | AP: 35.4 -> 33.3 | Hurt |
| add emergence simulation without DAQ (ablation) | DVIS baseline · OVIS (R50, 40K) | AP: 35.4 -> 32.3 | Hurt |
| combine emergence DAQ with emergence simulation (ablation) | DVIS baseline · OVIS (R50, 40K) | AP: 35.4 -> 36.4 | Improved |
| add disappearance DAQ (ablation) | DVIS-DAQ · OVIS (R50, 40K) | AP: 36.4 -> 36.6 | Improved |
| add disappearance simulation (ablation) | DVIS-DAQ · OVIS (R50, 40K) | AP: 36.6 -> 37.1 | Improved |
| use appearance features instead of positional features for emergence candidates (ablation) | emergence DAQ · OVIS (R50, 40K) | AP: 34.8 -> 36.4 | Improved |
| use candidate features as positional embedding instead of add or concat (ablation) | emergence DAQ · OVIS (R50, 40K) | AP: 36.0 -> 36.4 | Improved |
| initialize disappearance DAQ query feature from segmenter initial queries instead of a learnable embedding (ablation) | disappearance DAQ · OVIS (R50, 40K) | AP: 36.5 -> 37.1 | Improved |
| use very large emergence simulation threshold 0.5 (ablation) | emergence simulation · OVIS (R50, 40K) | AP: 35.6 -> 33.2 | Hurt |
| simulate disappearance on random frames instead of one frame or all frames (ablation) | disappearance simulation · OVIS (R50, 40K) | AP: 36.3 / 35.8 -> 37.1 | Improved |
| apply softmax over query dimension in disappearance tracker (ablation) | disappearance tracker · OVIS (R50, 40K) | AP: 36.6 -> 37.1 | Improved |
| pad offline refiner input with momentum-weighted appearance feature instead of zero (ablation) | DVIS-DAQ offline refiner · OVIS (R50) | AP: 39.0 -> 42.7 | Improved |
Closest papers by meaning
- Context-Aware Relative Object Queries To Unify Video Instance and Panoptic Segmentation (CVPR 2023 · similarity 0.88)
- Anchor Diffusion for Unsupervised Video Object Segmentation (ICCV 2019 · similarity 0.88)
- SyncVIS: Synchronized Video Instance Segmentation (NeurIPS 2024 · similarity 0.85)
- Learning Spatial-Semantic Features for Robust Video Object Segmentation (ICLR 2025 · similarity 0.85)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.85)
- Decoupled Motion Expression Video Segmentation (CVPR 2025 · similarity 0.84)
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.
