Research map / Video / Video segmentation and tracking
CompFeat: Comprehensive Feature Aggregation for Video Instance Segmentation
Yang Fu, Linjie Yang, Ding Liu, Thomas S. Huang, Humphrey Shi
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 temporal attention at frame level (ablation) | Baseline + MSCOCO · YouTube-VIS validation | AP: 24.1 -> 25.2 | Improved |
| Add spatial attention at frame level (ablation) | Baseline + MSCOCO · YouTube-VIS validation | AP: 24.1 -> 24.9 | Improved |
| Add temporal and spatial attention at frame level (ablation) | Baseline + MSCOCO · YouTube-VIS validation | AP: 24.1 -> 25.8 | Improved |
| Add temporal and spatial attention at object level (ablation) | Baseline + MSCOCO · YouTube-VIS validation | AP: 24.1 -> 26.1 | Improved |
| Add temporal and spatial attention at both frame and object levels (ablation) | Baseline + MSCOCO · YouTube-VIS validation | AP: 24.1 -> 27.5 | Improved |
| Add correlation-map tracking (ablation) | Baseline + MSCOCO · YouTube-VIS validation | AP: 24.1 -> 25.1 | Improved |
| Add frame-level dual attention to correlation-map tracking (ablation) | Baseline + CM · YouTube-VIS validation | AP: 25.1 -> 26.3 | Improved |
| Add object-level dual attention to correlation-map tracking (ablation) | Baseline + CM · YouTube-VIS validation | AP: 25.1 -> 26.7 | Improved |
| Combine correlation-map tracking with frame- and object-level dual attention (ablation) | Baseline + MSCOCO · YouTube-VIS validation | AP: 24.1 -> 28.4 | Improved |
| Use random instead of uniform training-frame sampling (ablation) | CompFeat · YouTube-VIS validation | AP: 27.8 -> 28.4 | Improved |
| Increase training frames from 2 to 3 with random sampling (ablation) | CompFeat · YouTube-VIS validation | AP: 28.4 -> 26.8 | Hurt |
| Use CompFeat rather than MaskTrack R-CNN | YouTube-VIS comparison · YouTube-VIS validation | AP: 30.3 -> 35.3 | Improved |
Later papers that compared with this method
2 later accepted papers report CompFeat as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- SeqFormer: Sequential Transformer for Video Instance Segmentation (ECCV 2022 · better on 1 of 1 shared result)
- Crossover Learning for Fast Online Video Instance Segmentation (ICCV 2021 · 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
- InsPro: Propagating Instance Query and Proposal for Online Video Instance Segmentation (NeurIPS 2022 · similarity 0.88)
- InstanceFormer: An Online Video Instance Segmentation Framework (AAAI 2023 · similarity 0.87)
- Video Instance Segmentation (ICCV 2019 · similarity 0.87)
- In Defense of Online Models for Video Instance Segmentation (ECCV 2022 · similarity 0.87)
- VideoCutLER: Surprisingly Simple Unsupervised Video Instance Segmentation (CVPR 2024 · similarity 0.87)
- Minimizing Labeled, Maximizing Unlabeled: An Image-Driven Approach for Video Instance Segmentation (CVPR 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.
