Research map / Video / Video segmentation and tracking
Semantic Flow for Fast and Accurate Scene Parsing
Xiangtai Li, Ansheng You, Zhen Zhu, Houlong Zhao, Maoke Yang, Kuiyuan Yang, Shaohua Tan, Yunhai Tong
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 bilinear upsampling with FAM in the FPN decoder (ablation) | FPN decoder (ResNet-18) · Cityscapes val | mIoU (%): 74.8 -> 77.2 | Improved |
| add FAM on top of FPN+PPM (ablation) | FPN+PPM (ResNet-18) · Cityscapes val | mIoU (%): 76.6 -> 78.7 | Improved |
| add FAM to all three stages F3, F4, F5 (ablation) | FPN+PPM · Cityscapes val | mIoU (%): 76.6 -> 78.3 | Improved |
| insert FAM only at the coarsest stage F5 versus F3 (ablation) | FPN+PPM · Cityscapes val | mIoU (%): 76.9 -> 77.5 | Improved |
| add FAM to ResNet-50 FPN+PPM (ablation) | ResNet-50 · Cityscapes val | mIoU (%): 76.8 -> 79.2 | Improved |
| add FAM to ResNet-101 FPN+PPM (ablation) | ResNet-101 · Cityscapes val | mIoU (%): 77.6 -> 79.8 | Improved |
| add FAM to light-weight DF2 (ablation) | DF2 · Cityscapes val | mIoU (%): 73.2 -> 75.8 | Improved |
| use PPM as the context head instead of Non-Local (ablation) | FAM decoder · Cityscapes val | mIoU (%): 76.8 -> 78.3 | Improved |
| use PPM instead of ASPP as context head (ablation) | FAM decoder · Cityscapes val | mIoU (%): 77.6 -> 78.3 | Improved |
| use deconvolution instead of bilinear upsampling inside FAM (ablation) | FAM · Cityscapes val | mIoU (%): 78.3 -> 77.9 | No clear effect |
| increase flow kernel size from 3x3 to 5x5 (ablation) | FAM · Cityscapes val | mIoU (%): 78.3 -> 78.1 | No clear effect |
| replace FlowNet-S style flow with FlowNet-C correlation (ablation) | FAM · Cityscapes val | mIoU (%): 77.5 -> 77.2 | Hurt |
Closest papers by meaning
- Sfnet: Faster and Accurate Semantic Segmentation Via Semantic Flow (IJCV 2023 · similarity 0.91)
- Semantic Flow: Learning Semantic Fields of Dynamic Scenes from Monocular Videos (ICLR 2024 · similarity 0.88)
- Pyramid Scene Parsing Network (CVPR 2017 · similarity 0.87)
- Temporal Pixel-Level Semantic Understanding Through the VSPW Dataset (TPAMI 2023 · similarity 0.87)
- VSPW: A Large-scale Dataset for Video Scene Parsing in the Wild (CVPR 2021 · similarity 0.87)
- Video Scene Parsing With Predictive Feature Learning (ICCV 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.
