Research map / Video / Video segmentation and tracking
Hybrid Instance-Aware Temporal Fusion 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 |
|---|---|---|---|
| Use hybrid instance-aware temporal fusion instead of existing online VIS approaches | Online video instance segmentation · YouTube-VIS-2019 val | AP: 41.3 vs 36.6 | Helped |
| Use hybrid instance-aware temporal fusion instead of the strongest listed ResNet-50 baseline | Video instance segmentation · YouTube-VIS-2021 val | AP: 34.2 -> 35.8 | Helped |
| Disable all inter-frame attention (ablation) | Proposed method · YouTube-VIS-2019 val | AP: 41.3 -> 36.9 | Hurt |
| Use only inter-frame p2c attention instead of the full attention design (ablation) | Proposed method · YouTube-VIS-2019 val | AP: 41.3 -> 38.8 | Hurt |
| Add c2c/c2p attention to p2c attention (ablation) | Proposed method · YouTube-VIS-2019 val | AP: 38.8 -> 40.7 | Helped |
| Add pair-wise matching to inter-frame attention (ablation) | Proposed method with p2c and c2c/c2p attention · YouTube-VIS-2019 val | AP: 40.7 -> 41.3 | No clear effect |
| Increase reference frames from 1 to 3 (ablation) | Proposed method · YouTube-VIS-2019 val | AP: 39.7 -> 41.3 | Helped |
| Increase reference frames from 3 to 4 (ablation) | Proposed method · YouTube-VIS-2019 val | AP: 41.3 -> 41.3 | No clear effect |
| Use 2 alternate attention layers rather than 1, 3, or 4 (ablation) | Proposed method · YouTube-VIS-2019 val | AP: 40.1 -> 41.3; 40.8; 40.2 | Helped |
| Vary the number of instance slots from 10 to 25 (ablation) | Proposed method · YouTube-VIS-2019 val | AP: 41.3; 41.2; 41.1; 41.3 | No clear effect |
Closest papers by meaning
- InstanceFormer: An Online Video Instance Segmentation Framework (AAAI 2023 · similarity 0.92)
- Spatial Feature Calibration and Temporal Fusion for Effective One-Stage Video Instance Segmentation (CVPR 2021 · similarity 0.90)
- Robust and Consistent Online Video Instance Segmentation via Instance Mask Propagation (AAAI 2025 · similarity 0.88)
- SyncVIS: Synchronized Video Instance Segmentation (NeurIPS 2024 · similarity 0.88)
- Minimizing Labeled, Maximizing Unlabeled: An Image-Driven Approach for Video Instance Segmentation (CVPR 2025 · similarity 0.88)
- Aligning Instance Brownian Bridge with Texts for Open-Vocabulary Video Instance Segmentation (AAAI 2025 · similarity 0.86)
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.