Research map / Detection and segmentation / CNN features
Sfnet: Faster and Accurate Semantic Segmentation Via Semantic Flow
Xiangtai Li, Jiangning Zhang, Yibo Yang, Guangliang Cheng, Kuiyuan Yang, Yunhai Tong, Dacheng Tao
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 |
|---|---|---|---|
| add Flow Alignment Module to FPN decoder (ablation) | FPN · Cityscapes val (ResNet-18) | mIoU (%): 74.8 -> 77.2 | Improved |
| add FAM to FPN + PPM (ablation) | FPN+PPM · Cityscapes val (ResNet-18) | mIoU (%): 76.6 -> 78.3 | Improved |
| use PPM instead of NL as context head (ablation) | FAM · Cityscapes val (ResNet-18) | mIoU (%): 76.8 -> 78.3 | Improved |
| use bilinear instead of deconvolution upsampling (ablation) | FAM · Cityscapes val (ResNet-18) | mIoU (%): 77.9 -> 78.3 | Improved |
| use bilinear instead of nearest upsampling (ablation) | FAM · Cityscapes val (ResNet-18) | mIoU (%): 78.2 -> 78.3 | No clear effect |
| use kernel size 3 instead of 1 (ablation) | FAM · Cityscapes val (ResNet-18) | mIoU (%): 77.8 -> 78.3 | Improved |
| use FAM instead of correlation flow (ablation) | FAM last stage · Cityscapes val (ResNet-18) | mIoU (%): 77.2 -> 77.5 | Improved |
| use FAM instead of DCN at three stages (ablation) | FPN+PPM · Cityscapes val (ResNet-18) | mIoU (%): 77.2 -> 78.3 | Improved |
| add dual flow to GD-FAM (ablation) | FPN+PPM · Cityscapes val (ResNet-18) | mIoU (%): 76.6 -> 77.8 | Improved |
| add attention and gate to dual flow (ablation) | GD-FAM · Cityscapes val (ResNet-18) | mIoU (%): 77.8 -> 78.3 | Improved |
| add uniform sampling, long training and coarse boosting (ablation) | SFNet-Lite · Cityscapes val (ResNet-18) | mIoU (%): 78.3 -> 79.7 | Improved |
| replace FAM with GD-FAM once (ablation) | FAM+PPM · Cityscapes val (ResNet-18) | FPS: 19.4 -> 24.5 | Improved |
Closest papers by meaning
- Semantic Flow for Fast and Accurate Scene Parsing (ECCV 2020 · similarity 0.91)
- DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs (TPAMI 2017 · similarity 0.89)
- Lightweight and Progressively-Scalable Networks for Semantic Segmentation (IJCV 2023 · similarity 0.88)
- BiSeNet V2: Bilateral Network with Guided Aggregation for Real-Time Semantic Segmentation (IJCV 2021 · similarity 0.88)
- Every Frame Counts: Joint Learning of Video Segmentation and Optical Flow (AAAI 2020 · similarity 0.88)
- UNIP: Rethinking Pre-trained Attention Patterns for Infrared Semantic Segmentation (ICLR 2025 · 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.
