Research map / Video / Video segmentation and tracking
Spatiotemporal Graph Neural Network based Mask Reconstruction for Video Object Segmentation
Daizong Liu, Shuangjie Xu, Xiao-Yang Liu, Zichuan Xu, Wei Wei, Pan Zhou
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 the motion model's predicted current-frame location instead of using the previous bounding box (ablation) | STG-Net proposal selection · DAVIS2017 test-dev | G_M: 53.2 -> 54.1; improvement of 0.9 | Improved |
| Add spatial graph reconstruction from all same-object proposals (ablation) | STG-Net · DAVIS2017 test-dev | G_M: 54.1 -> 58.4; improvement of 4.3 | Improved |
| Set spatial graph alpha to 0.7 rather than 0.5 (ablation) | STG-Net spatial graph · DAVIS2017 test-dev | G_M: 58.4 -> 59.2; improvement of 0.8 | Improved |
| Set mask-selection lambda1 to 0.4 rather than 0.5 (ablation) | STG-Net mask selection · DAVIS2017 test-dev | G_M: 59.2 -> 59.8; improvement of 0.6 | Improved |
| Add temporal graph retrieval of historic masks (ablation) | STG-Net · DAVIS2017 test-dev | G_M: 59.8 -> 61.9; improvement of 2.1 | Improved |
| Use 2 spatial graph updates rather than 1 (ablation) | STG-Net spatial graph · DAVIS2017 test-dev | G_M: 61.9 -> 63.1 | Improved |
| Increase spatial graph updates from 2 to 3 (ablation) | STG-Net spatial graph · DAVIS2017 test-dev | G_M: 63.1 -> 62.3 | Hurt |
| Add online learning on the first-frame annotation (ablation) | STG-Net · DAVIS2017 test-dev | G_M: 63.1 -> 64.3; improvement of 1.2 | Improved |
| Fine-tune on the augmented DAVIS2017 training set (ablation) | STG-Net · DAVIS2017 test-dev | G_M: 64.3 -> 66.5; improvement of 2.2 | Improved |
| Add YouTube-VOS pretraining (ablation) | STG-Net · DAVIS2017 test-dev | G_M: 66.5 -> 69.4; improvement of 2.9 | Improved |
| Use STG-Net rather than FEEL* | Video object segmentation · DAVIS2017 test-dev | G_M: 57.8 -> 63.1; 5.3% | Improved |
| Use STG-Net rather than AGSS | Video object segmentation · YouTube-VOS val | G_M: 71.3 -> 73.0; 1.7% | 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.
- RANet — RANet: Ranking Attention Network for Fast Video Object Segmentation (ICCV 2019) · 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
- Video Object Segmentation Using Space-Time Memory Networks (ICCV 2019 · similarity 0.89)
- Generating Masks From Boxes by Mining Spatio-Temporal Consistencies in Videos (ICCV 2021 · similarity 0.88)
- Video Object Segmentation with Episodic Graph Memory Networks (ECCV 2020 · similarity 0.88)
- Fast Video Object Segmentation by Reference-Guided Mask Propagation (CVPR 2018 · similarity 0.87)
- Spatiotemporal CNN for Video Object Segmentation (CVPR 2019 · similarity 0.87)
- Fast Video Object Segmentation via Dynamic Targeting Network (ICCV 2019 · similarity 0.87)
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.
