Research map / Video / Video segmentation and tracking
Tracking Anything with Decoupled Video Segmentation
Ho Kei Cheng, Seoung Wug Oh, Brian Price, Alexander Schwing, 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 |
|---|---|---|---|
| replace end-to-end Video-K-Net with decoupled DEVA propagation on same image model | Video-K-Net Swin-B · VIPSeg | VPQ-bar: 37.5 -> 46.4 | Improved |
| use decoupled propagation at frame level (VPQ-1) with Swin-B Video-K-Net online | Video-K-Net Swin-B · VIPSeg | VPQ-1: 49.8 -> 48.2 | Hurt |
| use bi-directional propagation instead of short-term association schemes (mask IoU) (ablation) | DEVA · VIPSeg | VPQ-bar: 27.6 -> 36.4 | Improved |
| merge propagated and consensus results instead of trusting image segmentation only (ablation) | DEVA · VIPSeg | VPQ-bar: 33.2 -> 36.4 | Improved |
| add spatial alignment in in-clip consensus (ablation) | DEVA · VIPSeg | VPQ-bar: 32.8 -> 38.3 | Improved |
| increase clip size from 1 to 3 in in-clip consensus (ablation) | DEVA · VIPSeg | VPQ-bar: 36.4 -> 38.3 | Improved |
| increase clip size from 3 to 5 (ablation) | DEVA · VIPSeg | VPQ-bar: 38.3 -> 38.3 | No clear effect |
| merge consensus every 7 frames instead of every 5 (ablation) | DEVA · VIPSeg | VPQ-bar: 38.3 -> 35.7 | Hurt |
| swap in EntitySeg instead of Mask2Former as image model | DEVA semi-online · BURST test | OWTA uncommon: 44.1 -> 53.3 | Improved |
| use better image backbone (Swin-B instead of R50) without changing propagation | DEVA with Mask2Former · VIPSeg | VPQ-bar: 38.3 -> 52.2 | Improved |
Closest papers by meaning
- Tracking and Segmenting Anything in Any Modality (AAAI 2026 · similarity 0.89)
- STEP: Segmenting and Tracking Every Pixel (NeurIPS 2021 · similarity 0.88)
- DVIS++: Improved Decoupled Framework for Universal Video Segmentation (TPAMI 2025 · similarity 0.87)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.85)
- Segment Any Motion in Videos (CVPR 2025 · similarity 0.85)
- Matching Anything by Segmenting Anything (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.
