Research map / Video / Video segmentation and tracking
InsPro: Propagating Instance Query and Proposal for Online Video Instance Segmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Propagate instance queries and proposals across frames rather than processing each frame independently (ablation) | InsPro · YouTube-VIS 2019 | AP: 24.0 -> 36.3 | Helped |
| Add temporally consistent matching during training (ablation) | InsPro with query and proposal propagation · YouTube-VIS 2019 | AP: 36.3 -> 37.4 | Helped |
| Propagate both queries and proposals instead of only one (ablation) | InsPro with TCM · YouTube-VIS 2019 | AP: 36.7, 36.6 -> 37.4 | Helped |
| Add box deduplication loss (ablation) | InsPro · YouTube-VIS 2019 | AP: 37.4 -> 38.4 | Helped |
| Use intra-query attention with a longer feature bank, from T=1 to T=18 (ablation) | InsPro · YouTube-VIS 2019 | AP: 38.4 -> 40.2 | Helped |
| Increase feature bank length beyond T=18 (ablation) | InsPro · YouTube-VIS 2019 | AP: 40.2 -> 40.1 | No clear effect |
| Replace tracking-by-detection with query-proposal temporal propagation (ablation) | VIS with the same instance segmentation baseline · YouTube-VIS 2019 | AP: 31.5 -> 37.4 | Helped |
| Use one united query set instead of separate track and detect query sets (ablation) | InsPro · YouTube-VIS 2019 | AP: 37.4 -> 38.4 | Helped |
| Increase the InsPro-lite key-frame interval from K=1 to K=15 (ablation) | InsPro-lite · YouTube-VIS 2019 | AP: 40.2 -> 37.5 | Hurt |
| Add temporally consistent matching to temporal propagation (ablation) | InsPro-VID · ImageNet VID | AP: 29.2 -> 33.2 | Helped |
| Increase BDL beta from 0 to 0.15 (ablation) | InsPro-VID · ImageNet VID | AP: 33.2 -> 39.5 | Helped |
| Increase intra-query feature bank length from T=20 to T=100 (ablation) | InsPro-VID · ImageNet VID | AP: 39.8 -> 42.2 | Helped |
Closest papers by meaning
- SyncVIS: Synchronized Video Instance Segmentation (NeurIPS 2024 · similarity 0.90)
- OpenVIS: Open-vocabulary Video Instance Segmentation (AAAI 2025 · similarity 0.89)
- Robust and Consistent Online Video Instance Segmentation via Instance Mask Propagation (AAAI 2025 · similarity 0.89)
- Minimizing Labeled, Maximizing Unlabeled: An Image-Driven Approach for Video Instance Segmentation (CVPR 2025 · similarity 0.89)
- Aligning Instance Brownian Bridge with Texts for Open-Vocabulary Video Instance Segmentation (AAAI 2025 · similarity 0.88)
- Decoupled Motion Expression Video Segmentation (CVPR 2025 · 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.