Research map / Video / Video segmentation and tracking
Memory Selection Network for Video Propagation
Ruizheng Wu, Huaijia Lin, Xiaojuan Qi, Jiaya Jia
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 memory selection network (MSN) to select the guidance frame | TPN · YouTube-VOS val | Overall (%): 63.0 -> 64.5 (+1.5) | Improved |
| add MSN to select one memorized frame | STM-1 · YouTube-VOS val | Overall (%): 69.9 -> 71.1 (+1.2) | Improved |
| add MSN to TPN | TPN · DAVIS-2016 val | J (%): 75.8 -> 76.8 | Improved |
| add MSN to STM-1 | STM-1 · DAVIS-2017 val | J (%): 69.6 -> 71.4 | Improved |
| add MSN to TPN for colorization | TPN · DAVIS-2016 grayscale video colorization | PSNR: 28.25 -> 28.57 | Improved |
| select guidance with trained MSN instead of VGG feature distance (ablation) | TPN · YouTube-VOS val (selection strategy, TPN) | Overall (%): 63.03 -> 64.5 | Improved |
| select guidance with MSN instead of fixed time gap 5 (ablation) | TPN · YouTube-VOS val (selection strategy, TPN) | Overall (%): 63.13 -> 64.5 | Improved |
| increase fixed time step gap from 1 to 20 (ablation) | TPN · YouTube-VOS val (selection strategy, TPN) | Overall (%): 63.04 -> 55.61 | Hurt |
| add MSN-based top-3 selection to the 3-frame ensemble (ablation) | TPN · YouTube-VOS val (ensemble, TPN) | Overall (%): 65.27 -> 65.8 | Improved |
| add MSN-based top-5 selection to the 5-frame ensemble (ablation) | TPN · YouTube-VOS val (ensemble, TPN) | Overall (%): 65.6 -> 66.1 | Improved |
| increase ensemble size from 1 to 5 frames with MSN (ablation) | TPN + MSN · YouTube-VOS val (ensemble, TPN) | Overall (%): 64.54 -> 66.1 | Improved |
| select guidance with ground-truth oracle instead of learned MSN (ablation) | TPN + MSN · YouTube-VOS val | Overall (%): 64.5 -> 75.3 | Improved |
Closest papers by meaning
- Video Propagation Networks (CVPR 2017 · similarity 0.89)
- ColorMNet: A Memory-based Deep Spatial-Temporal Feature Propagation Network for Video Colorization (ECCV 2024 · similarity 0.87)
- Switchable Temporal Propagation Network (ECCV 2018 · similarity 0.86)
- Learning Video Object Segmentation With Visual Memory (ICCV 2017 · similarity 0.85)
- Frame Context Packing and Drift Prevention in Next-Frame-Prediction Video Diffusion Models (NeurIPS 2025 · similarity 0.83)
- CoRDS: Coreset-Based Representative and Diverse Selection for Streaming Video Understanding (NeurIPS 2026 · similarity 0.82)
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.
