Research map / Video / Video segmentation and tracking
TapLab: A Fast Framework for Semantic Video Segmentation Tapping Into Compressed-Domain Knowledge
Junyi Feng, Songyuan Li, Xi Li, Fei Qing Wu, Qi Tian, Ming–Hsuan Yang, Haibin Ling
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 |
|---|---|---|---|
| replace per-frame inference with fast feature warping (FFW) via motion vectors | BL2 · Cityscapes val | mIoU: 73.6 -> 64.4 | Hurt |
| replace per-frame inference with FFW | BL2 · Cityscapes val | FPS: 29.3 -> 160.4 | Improved |
| use FFW motion vectors instead of linear interpolation (ablation) | BL2 · Cityscapes val | mIoU: 42.1 -> 64.4 | Improved |
| use motion vectors (FFW) instead of FlowNet2 optical flow (ablation) | BL2 · Cityscapes val | mIoU: 62.1 -> 64.4 | No clear effect |
| add residual-guided correction (RGC) to FFW (ablation) | BL2 · Cityscapes val | mIoU: 64.4 -> 68.2 | Improved |
| add residual-guided frame selection (RGFS) to FFW+RGC (ablation) | BL2 · Cityscapes val | mIoU: 68.2 -> 70.6 | Improved |
| add RGC and RGFS to FFW (ablation) | BL3 · Cityscapes val | mIoU: 67.1 -> 74.5 | Improved |
| add RGC and RGFS to FFW (ablation) | BL2 · Cityscapes val | FPS: 160.4 -> 99.8 | Hurt |
| use larger overlapping RGC regions (512x512, stride 256) instead of 256x256 (ablation) | BL1+FFW+RGC · Cityscapes val | mIoU: 62.4 -> 63.4 | Improved |
| apply RGC and RGFS under larger GOP number 12 instead of 3 (ablation) | BL2+FFW · Cityscapes val | mIoU gain: +0.6 -> +6.2 | Improved |
| use PWC-Net optical flow with RGC and RGFS instead of motion vectors (ablation) | BL2 · Cityscapes val | mIoU: 70.6 -> 68.9 | Hurt |
| combine warped and CNN features (intermediate alpha) instead of CNN only (alpha=1) (ablation) | RGC · Cityscapes val | mIoU | Improved |
Closest papers by meaning
- Efficient Semantic Segmentation by Altering Resolutions for Compressed Videos (CVPR 2023 · similarity 0.88)
- Low-Latency Video Semantic Segmentation (CVPR 2018 · similarity 0.88)
- Exploring Temporal Feature Correlation for Efficient and Stable Video Semantic Segmentation (AAAI 2024 · similarity 0.87)
- Learning Spatial-Semantic Features for Robust Video Object Segmentation (ICLR 2025 · similarity 0.86)
- VidSeg: Training-free Video Semantic Segmentation based on Diffusion Models (CVPR 2025 · similarity 0.86)
- Exploiting Temporal State Space Sharing for Video Semantic Segmentation (CVPR 2025 · similarity 0.86)
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.
