Research map / Detection and segmentation / CNN features
Decoupling Features in Hierarchical Propagation for Video Object Segmentation
Zongxin Yang, 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 |
|---|---|---|---|
| decouple visual and ID embedding propagation (dual-branch) (ablation) | DeAOT-S · YouTube-VOS 2018 val (DeAOT-S ablation) | J&F: 81.5 -> 82.5 | Improved |
| double channel dimensions of coupled propagation to 512 (ablation) | AOT-style coupling · YouTube-VOS 2018 val (DeAOT-S ablation) | J&F: 81.5 -> 82.0 | Improved |
| replace LSTT with GPM module (ablation) | DeAOT-S · YouTube-VOS 2018 val (DeAOT-S ablation) | J&F: 80.3 -> 82.5 | Improved |
| reduce attention heads from 8 to 1 (ablation) | AOT · YouTube-VOS 2018 val (DeAOT-S ablation) | J&F: 80.3 -> 79.6 | Hurt |
| reduce attention heads from 8 to 1 (ablation) | DeAOT · YouTube-VOS 2018 val (DeAOT-S ablation) | J&F: 82.5 -> 82.5 | No clear effect |
| add ID embedding to long/short-term attention maps (ablation) | DeAOT-S · YouTube-VOS 2018 val (DeAOT-S ablation) | J&F: 82.5 -> 82.1 | Hurt |
| use ID embedding as positional embedding in self-propagation (ablation) | DeAOT-S · YouTube-VOS 2018 val (DeAOT-S ablation) | J&F: 82.2 -> 82.5 | Improved |
| add depthwise convolution in GPM (ks=5) (ablation) | DeAOT-S · YouTube-VOS 2018 val (DeAOT-S ablation) | J&F: 81.1 -> 82.5 | Improved |
| kernel size 5 versus 3 or 9 (ablation) | DeAOT-S · YouTube-VOS 2018 val (DeAOT-S ablation) | J&F: 82.2 / 82.4 / 82.5 | No clear effect |
| replace AOT with DeAOT | R50-AOT-L · YouTube-VOS 2018 val | J&F: 84.1 -> 86.0 | Improved |
| replace AOT with DeAOT | AOT-S · DAVIS 2017 val | J&F: 81.3 -> 80.8 | Hurt |
| replace AOT with DeAOT | AOT-T · speed | fps: 41.0 -> 53.4 | Improved |
Closest papers by meaning
- Associating Objects with Transformers for Video Object Segmentation (NeurIPS 2021 · similarity 0.89)
- Scalable Video Object Segmentation With Identification Mechanism (TPAMI 2024 · similarity 0.89)
- MHP-VOS: Multiple Hypotheses Propagation for Video Object Segmentation (CVPR 2019 · similarity 0.88)
- Siamese Network with Interactive Transformer for Video Object Segmentation (AAAI 2022 · similarity 0.87)
- Breaking the "Object" in Video Object Segmentation (CVPR 2023 · similarity 0.86)
- Referred by Multi-Modality: A Unified Temporal Transformer for Video Object Segmentation (AAAI 2024 · similarity 0.85)
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.
