Research map / Video / Video segmentation and tracking
Tube-Link: A Flexible Cross Tube Framework for Universal Video Segmentation
Xiangtai Li, Haobo Yuan, Wenwei Zhang, Guangliang Cheng, Jiangmiao Pang, Chen Change Loy
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 |
|---|---|---|---|
| use tube matching (2-frame subclip) instead of frame matching in the Mask2Former-VIS+ baseline (ablation) | Mask2Former-VIS+ · VIPSeg val (ResNet50) | VPQ: 32.4 -> 34.5 | Improved |
| add cross-tube temporal contrastive learning (TCL) (ablation) | Tube-Link · VIPSeg val (ResNet50) | VPQ: 34.5 -> 36.8 | Improved |
| add cross-tube linking (CTL) on top of TCL (ablation) | Tube-Link · VIPSeg val (ResNet50) | VPQ: 36.8 -> 37.5 | Improved |
| replace tube-mask association targets with single-frame GT masks (ablation) | Tube-Link TCL · VIPSeg val (ResNet50) | VPQ: 37.5 -> 35.6 | Hurt |
| use global query matching rather than dense or sparse query matching for contrastive learning (ablation) | Tube-Link TCL · VIPSeg val (ResNet50) | VPQ: 34.5 -> 37.5 vs sparse query; 30.2 -> 37.5 vs dense | Improved |
| increase training subclip size from 1 to 2 (ablation) | Tube-Link · VIPSeg val (ResNet50) | VPQ: 35.6 -> 37.5 | Improved |
| increase training subclip size from 2 to 3 or add frame overlap (ablation) | Tube-Link · VIPSeg val (ResNet50) | VPQ: 37.5 -> 37.0 (T=3); 37.3 (overlap) | No clear effect |
| enlarge inference window from 2 to 6 (trained with 2) (ablation) | Tube-Link · VIPSeg val (ResNet50) | VPQ: 37.5 -> 39.2 | Improved |
| enlarge inference window further to 8 (ablation) | Tube-Link · VIPSeg val (ResNet50) | VPQ: 39.2 -> 38.5 | Hurt |
| use learned tube embedding for tracking instead of extra tracker, RoI features or query embedding (ablation) | Tube-Link · VIPSeg val (ResNet50) | VPQ: 36.6 -> 37.5 vs extra tracker | Improved |
| feed two-frame subclips to pretrained Min-VIS and Video K-Net without retraining | Min-VIS / Video K-Net · YouTube-VIS-2019 / VIPSeg | mAP / VPQ: 47.4 -> 48.8 (Min-VIS YTVIS19); 26.1 -> 27.6 (Video K-Net VIPSeg) | Improved |
| Tube-Link versus TubeFormer on VSPW semantic segmentation | Tube-Link Swin-large · VSPW-VSS val | mIoU: 59.7 vs 63.2 (TubeFormer) | Hurt |
Later papers that compared with this method
2 later accepted papers report Tube-Link as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- General and Task-Oriented Video Segmentation (ECCV 2024 · better on 3 of 3 shared results)
- DVIS: Decoupled Video Instance Segmentation Framework (ICCV 2023 · better on 1 of 1 shared result)
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.
- Video K-Net — Video K-Net: A Simple, Strong, and Unified Baseline for Video Segmentation (CVPR 2022) · better on 5 of 6 shared results
- Min-VIS — MinVIS: A Minimal Video Instance Segmentation Framework without Video-based Training (NeurIPS 2022) · better on 1 of 1 shared result
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
- TubeFormer-DeepLab: Video Mask Transformer (CVPR 2022 · similarity 0.87)
- Video K-Net: A Simple, Strong, and Unified Baseline for Video Segmentation (CVPR 2022 · similarity 0.85)
- Object Detection in Videos by High Quality Object Linking (TPAMI 2019 · similarity 0.85)
- Learning Spatial-Semantic Features for Robust Video Object Segmentation (ICLR 2025 · similarity 0.85)
- Minimizing Labeled, Maximizing Unlabeled: An Image-Driven Approach for Video Instance Segmentation (CVPR 2025 · similarity 0.84)
- VideoCutLER: Surprisingly Simple Unsupervised Video Instance Segmentation (CVPR 2024 · 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.
