Research map / Video / Video segmentation and tracking
Scalable Video Object Segmentation With Identification Mechanism
Zongxin Yang, Jiaxu Miao, Yunchao Wei, Wenguan Wang, Xiaohan Wang, Yi Yang
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 mechanism with multi-object association (n_id=10) versus n_id=1 with post-ensemble (ablation) | AOT · YouTube-VOS (ablation) | J&F: 78.7 -> 80.3 | Improved |
| set identity bank size n_id larger than 10 (ablation) | AOT · YouTube-VOS (ablation) | J&F | Hurt |
| add layer-wise ID-based attention and scalable supervision (AOST over AOT-L) | AOT-L · YouTube-VOS 2018 Val | J&F: 83.8 -> 84.4 | Improved |
| add layer-wise ID-based attention and scalable supervision (AOST over AOT-L) | AOT-L · VOSW | J&F: 74.8 -> 75.5 | Improved |
| reduce inference depth of AOST to n_l'=1 (ablation) | AOST · YouTube-VOS 2018 Val | J&F: 84.4 -> 80.6 | Hurt |
| reduce inference depth of AOST to n_l'=2 (ablation) | AOST · DAVIS 2017 Test | J&F: 79.9 -> 78.1 | Mixed |
| set alpha=infinity in scalable supervision (ablation) | R50-AOST n_l'=3 · YouTube-VOS 2018 Val | J&F: 85.0 -> 84.8 | Mixed |
| use all-frame 30FPS videos at test time | R50-AOST alpha=inf · YouTube-VOS 2018 Val | J&F: 84.8 -> 85.4 | Improved |
| replace ResNet-50 with Swin-B backbone | AOST · YouTube-VOS 2018 Val (all-frame) | J&F: 85.4 -> 85.8 | Improved |
| replace ResNet-50 with larger Swin-B backbone in cross-dataset evaluation | AOT-L · VOSW | J&F: 74.8 -> 74.6 | Hurt |
| add identification mechanism (AOT-L over STCN) for multi-object tracking | STCN · VOSW | J&F: 73.1 -> 74.8 | Improved |
| AOT-S with identification mechanism versus STCN | STCN · YouTube-VOS 2018 Val | FPS: 8.4 -> 27.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- XMem — XMem: Long-Term Video Object Segmentation with an Atkinson-Shiffrin Memory Model (ECCV 2022) · better on 2 of 2 shared results
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
- Associating Objects with Transformers for Video Object Segmentation (NeurIPS 2021 · similarity 0.94)
- MHP-VOS: Multiple Hypotheses Propagation for Video Object Segmentation (CVPR 2019 · similarity 0.90)
- Breaking the "Object" in Video Object Segmentation (CVPR 2023 · similarity 0.90)
- Siamese Network with Interactive Transformer for Video Object Segmentation (AAAI 2022 · similarity 0.90)
- Learning Fast and Robust Target Models for Video Object Segmentation (CVPR 2020 · similarity 0.89)
- Point-VOS: Pointing Up Video Object Segmentation (CVPR 2024 · similarity 0.89)
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.
