Research map / Video / Video segmentation and tracking
Instance As Identity: A Generic Online Paradigm for Video Instance Segmentation
Feng Zhu, Zongxin Yang, Xin Yu, Yi Yang, Yunchao Wei
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 identification and association modules together (ablation) | CondInst image model · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 24.0 -> 37.9 | Improved |
| add identification module without association (ablation) | CondInst image model · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 24.0 -> 12.6 | Hurt |
| add association module without identification (ablation) | CondInst image model · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 24.0 -> 24.5 | No clear effect |
| add global memory to HAB (ablation) | IAI HAB · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 5.0 mAP | Improved |
| add local memory to HAB (ablation) | IAI HAB · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 3.0 mAP | Improved |
| add classification projector to HAB (ablation) | IAI HAB · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 36.9 -> 37.9 | Improved |
| use IAI paradigm instead of tracking-by-detection with Mask R-CNN | Mask R-CNN · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 30.3 -> 31.7 | Improved |
| use IAI paradigm instead of tracking-by-detection with CondInst | CondInst · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 32.1 -> 37.9 | Improved |
| use focal loss instead of cross-entropy for ID loss (ablation) | IAI+CondInst · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 36.4 -> 37.9 | Improved |
| use 3-conv instead of 2-conv ID head (ablation) | IAI+CondInst · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 37.9 -> 38.1 | No clear effect |
| use 4-conv instead of 2-conv ID head (ablation) | IAI+CondInst · YouTube-VIS-2019 val (ablation, R50 no aug) | mAP: 37.9 -> 35.3 | Hurt |
| pretrain on YouTube-VIS 2019 before OVIS training | IAI+CondInst · OVIS val | mAP: 18.5 -> 20.6 | Improved |
Closest papers by meaning
- InsPro: Propagating Instance Query and Proposal for Online Video Instance Segmentation (NeurIPS 2022 · similarity 0.90)
- In Defense of Online Models for Video Instance Segmentation (ECCV 2022 · similarity 0.89)
- InstanceFormer: An Online Video Instance Segmentation Framework (AAAI 2023 · similarity 0.89)
- Robust and Consistent Online Video Instance Segmentation via Instance Mask Propagation (AAAI 2025 · similarity 0.89)
- Video Instance Segmentation (ICCV 2019 · similarity 0.89)
- VISAGE: Video Instance Segmentation with Appearance-Guided Enhancement (ECCV 2024 · similarity 0.88)
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.
