Research map / Video / Video segmentation and tracking
Dynamic Video Segmentation Network
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 DVSNet adaptive routing with expected-confidence thresholding | DeepLab-Fast on Cityscapes · Cityscapes validation, balanced mode | fps: 5.6 -> 19.8 | Helped |
| Add DVSNet adaptive routing with expected-confidence thresholding | DeepLab-Fast on Cityscapes · Cityscapes validation, balanced mode | mIoU (%): 73.5 -> 70.4 | Hurt |
| Lower the expected-confidence threshold | DVSNet · Cityscapes | fps | Helped |
| Lower the expected-confidence threshold | DVSNet · Cityscapes | mIoU | Hurt |
| Use expected confidence score instead of fixed key-frame scheduling (ablation) | DVSNet with DeepLab-Fast and FlowNet2-s · Cityscapes | mIoU | Helped |
| Use frame difference as the scheduling decision metric instead of expected confidence score (ablation) | DVSNet with DeepLab-Fast and FlowNet2-s · Cityscapes | mIoU | Hurt |
| Use conv6 feature maps as DN input rather than key-frame + flow inputs (ablation) | DVSNet decision network · DN confidence-score prediction | error rate (%): 2.20 -> 1.74 | Helped |
| Use conv6 feature maps as DN input rather than conv4 feature maps (ablation) | DVSNet decision network · DN confidence-score prediction | error rate (%): 1.94 -> 1.74 | Helped |
| Divide frames into 2 × 2 regions (ablation) | DVSNet · Cityscapes frame-region routing | spatial warping / segmentation ratio: 34,013 / 5,987 = 5.68 | Helped |
| Use no overlap between adjacent frame regions (ablation) | DVSNet (DeepLab-Fast, FlowNet2-s, t = 92%) · Cityscapes | mIoU (%): 65.5 | Hurt |
| Add 64-pixel overlap between adjacent frame regions (ablation) | DVSNet (DeepLab-Fast, FlowNet2-s, t = 92%) · Cityscapes | mIoU (%): 70% | Helped |
Closest papers by meaning
- Temporally Distributed Networks for Fast Video Semantic Segmentation (CVPR 2020 · similarity 0.90)
- Low-Latency Video Semantic Segmentation (CVPR 2018 · similarity 0.89)
- DFANet: Deep Feature Aggregation for Real-Time Semantic Segmentation (CVPR 2019 · similarity 0.88)
- Learning Dynamic Network Using a Reuse Gate Function in Semi-Supervised Video Object Segmentation (CVPR 2021 · similarity 0.87)
- Motion-Guided Cascaded Refinement Network for Video Object Segmentation (CVPR 2018 · similarity 0.87)
- Budget-Aware Deep Semantic Video Segmentation (CVPR 2017 · 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.