Research map / Detection and segmentation / CNN features
Asymmetric Non-Local Neural Networks for Semantic Segmentation
Zhen Zhu, Mengde Xu, Song Bai, Tengteng Huang, Xiang Bai
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 non-local block before classifier (ablation) | FCN-like ResNet-101 baseline · Cityscapes val | mIoU: 75.8 -> 78.4 | Improved |
| replace non-local block with APNB (ablation) | ResNet-101 + NB · Cityscapes val | mIoU: 78.4 -> 78.6 | No clear effect |
| replace non-local block with APNB for efficiency | non-local block · Cityscapes 256x128 input | GFLOPs: 601.4 -> 43.5 | Improved |
| replace non-local block with APNB for efficiency | non-local block · Cityscapes 256x128 input | GPU memory (MB): 7797 -> 277 | Improved |
| add common fusion of Stage4 and Stage5 (ablation) | ResNet-101 baseline · Cityscapes val | mIoU: 75.8 -> 76.5 | Improved |
| replace common fusion with fusion non-local block (ablation) | ResNet-101 + fusion · Cityscapes val | mIoU: 76.5 -> 77.3 | Improved |
| use AFNB instead of FNB (ablation) | ResNet-101 + fusion · Cityscapes val | mIoU: 77.3 -> 77.1 | No clear effect |
| add common fusion to non-local model (ablation) | ResNet-101 + NB · Cityscapes val | mIoU: 78.4 -> 79.0 | Improved |
| add AFNB to APNB model (ablation) | ResNet-101 + APNB · Cityscapes val | mIoU: 78.6 -> 79.9 | Improved |
| use pyramid average sampling instead of plain average sampling (ablation) | APNB · Cityscapes val | mIoU: 78.4 -> 79.3 | Improved |
| use average sampling instead of max sampling (ablation) | APNB · Cityscapes val | mIoU: 78.1 -> 78.4 | Improved |
| increase anchor points from 50 to 225 with pyramid sizes (1,4,8,12) (ablation) | APNB · Cityscapes val | mIoU: 79.3 -> 80.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PSANet — PSANet: Point-wise Spatial Attention Network for Scene Parsing (ECCV 2018) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs (TPAMI 2017 · similarity 0.87)
- BiSeNet V2: Bilateral Network with Guided Aggregation for Real-Time Semantic Segmentation (IJCV 2021 · similarity 0.86)
- Lightweight and Progressively-Scalable Networks for Semantic Segmentation (IJCV 2023 · similarity 0.86)
- Disentangled Non-local Neural Networks (ECCV 2020 · similarity 0.86)
- ALGM: Adaptive Local-then-Global Token Merging for Efficient Semantic Segmentation with Plain Vision Transformers (CVPR 2024 · similarity 0.85)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 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.
