Research map / Detection and segmentation / CNN features
Improving Semantic Segmentation via Video Propagation and Label Relaxation
Check 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 Mapillary pre-training (ablation) | Cityscapes segmentation baseline · Cityscapes val | mIoU: 76.60 -> 78.32 | Helped |
| Add class uniform sampling for fine and coarse annotations (ablation) | Cityscapes model with Mapillary pre-training · Cityscapes val | mIoU: 78.32 -> 79.46 | Helped |
| Use joint image-label propagation instead of label propagation (ablation) | Video-prediction augmentation · Cityscapes val, ±1 propagation | mIoU: 79.79 -> 80.26 | Helped |
| Use video reconstruction instead of video prediction (ablation) | Joint image-label propagation · Cityscapes val, ±1 propagation | mIoU: 80.26 -> 80.54 | Helped |
| Add joint propagation with video reconstruction (ablation) | Cityscapes baseline · Cityscapes val, ±1 propagation | mIoU: 79.46 -> 80.54 | Helped |
| Add boundary label relaxation (ablation) | Video-reconstruction joint-propagation training at ±3 · Cityscapes val | mIoU: 80.54 -> 81.35 | Helped |
| Add boundary label relaxation without propagation (ablation) | Cityscapes segmentation baseline · Cityscapes val, no propagation | mIoU: 79.46 -> 80.85 | Helped |
| Use accumulated synthesized samples from ±1 through ±3 instead of only ±3 (ablation) | Joint propagation with boundary label relaxation · Cityscapes val | mIoU: 81.35 -> 81.12 | Hurt |
| Use FlowNet2 optical flow instead of learned video-reconstruction motion vectors (ablation) | Propagation-based training augmentation · Cityscapes val | mIoU | Hurt |
| Train with synthesized samples and boundary label relaxation (ablation) | CamVid model · CamVid test | mIoU: 79.8 -> 81.7 | Helped |
Closest papers by meaning
- Semantic Video Segmentation by Gated Recurrent Flow Propagation (CVPR 2018 · similarity 0.90)
- Low-Latency Video Semantic Segmentation (CVPR 2018 · similarity 0.89)
- Mask Propagation for Efficient Video Semantic Segmentation (NeurIPS 2023 · similarity 0.89)
- VidSeg: Training-free Video Semantic Segmentation based on Diffusion Models (CVPR 2025 · similarity 0.87)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 2024 · similarity 0.87)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · 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.