Research map / Video / Video segmentation and tracking
Efficient Semantic Segmentation by Altering Resolutions for Compressed Videos
Yubin Hu, Yuze He, Yanghao Li, Jisheng Li, Yuxing Han, Jiangtao Wen, Yong-Jin Liu
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 |
|---|---|---|---|
| Alter input resolution between HR keyframes and LR non-keyframes, with CReFF and FST | PSPNet18 and BiseNet18 · CamVid test and Cityscapes valid | mIoU (%): 69.43 -> 70.48; 71.57 -> 70.38; 69.00 -> 69.03; 70.09 -> 70.57 | Improved |
| Use AR-Seg with lower-resolution non-keyframes | PSPNet18 and BiseNet18 · CamVid test and Cityscapes valid | GFLOPs: 309.02 -> 101.98; 58.83 -> 18.96; 560.97 -> 177.44; 178.96 -> 57.00 | Improved |
| Remove CReFF (ablation) | AR-Seg with PSPNet18 · CamVid test; L=12 | mIoU (%): 70.48 -> 67.14 | Hurt |
| Use motion-vector warping without local attention (ablation) | AR-Seg with PSPNet18 · CamVid test; L=12 | mIoU (%): 70.48 -> 57.64 | Hurt |
| Use local attention without motion-vector warping (ablation) | AR-Seg with PSPNet18 · CamVid test; L=12 | mIoU (%): 70.48 -> 67.93 | Hurt |
| Replace 7x7 local attention with 3x3 local attention (ablation) | AR-Seg with PSPNet18 · CamVid test; L=12 | mIoU (%): 70.48 -> 70.30 | Hurt |
| Replace 7x7 local attention with 11x11 local attention (ablation) | AR-Seg with PSPNet18 · CamVid test; L=12 | GFLOPs: 101.98 -> 107.32 | Hurt |
| Replace grouped local-attention encoders with non-grouped encoders (ablation) | AR-Seg with PSPNet18 · CamVid test; L=12 | GFLOPs: 101.98 -> 170.96 | Hurt |
| Replace local attention with global attention (ablation) | AR-Seg with PSPNet18 · CamVid test; L=12 | mIoU (%): 70.48 -> 67.11 | Hurt |
| Replace local attention with a one-layer convolution (ablation) | AR-Seg with PSPNet18 · CamVid test; L=12 | GFLOPs: 101.98 -> 143.63 | Hurt |
| Remove the direct-connection path from CReFF (ablation) | AR-Seg with PSPNet18 · CamVid test; L=12 | mIoU (%): 70.48 -> 69.14 | Hurt |
| Move CReFF before the task sub-network instead of before the final 1x1 convolution (ablation) | AR-Seg with PSPNet18 · CamVid test; L=12 | mIoU (%): 70.48 -> 68.60 | Hurt |
Closest papers by meaning
- TapLab: A Fast Framework for Semantic Video Segmentation Tapping Into Compressed-Domain Knowledge (TPAMI 2020 · similarity 0.88)
- Mask Propagation for Efficient Video Semantic Segmentation (NeurIPS 2023 · similarity 0.87)
- VidSeg: Training-free Video Semantic Segmentation based on Diffusion Models (CVPR 2025 · similarity 0.87)
- Seg-ReSearch: Segmentation with Interleaved Reasoning and External Search (ICML 2026 · similarity 0.86)
- Exploring Temporal Feature Correlation for Efficient and Stable Video Semantic Segmentation (AAAI 2024 · similarity 0.86)
- Vanishing-Point-Guided Video Semantic Segmentation of Driving Scenes (CVPR 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.
