Research map / Detection and segmentation / CNN features
Adaptive Context Network for Scene Parsing
Jun Fu, Jing Liu, Yuhang Wang, Yong Li, Yongjun Bao, Jinhui Tang, Hanqing Lu
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 global context module (gated by pixel-feature similarity) instead of plain global pooling sum (ablation) | Res-50 dilated FCN (output 1/16) · Cityscapes val | mIoU: 71.24 -> 72.45 | Improved |
| same, with output stride 1/8 (ablation) | Res-50 dilated FCN (output 1/8) · Cityscapes val | mIoU: 72.77 -> 74.50 | Improved |
| add global context module to baseline (ablation) | Res-50 (output 1/16) · Cityscapes val | mIoU: 69.15 -> 72.45 | Improved |
| set delta to 5 rather than 10 (ablation) | Global context module · Cityscapes val | mIoU: 71.87 -> 72.45 | Improved |
| set delta to 5 rather than 2 (ablation) | Global context module · Cityscapes val | mIoU: 72.36 -> 72.45 | No clear effect |
| add ungated local context from block-2 (ablation) | Res-50+GCM · Cityscapes val | mIoU: 72.45 -> 73.48 | Improved |
| gate local context with reversed global coefficient (LCM) (ablation) | Res-50+GCM+LC · Cityscapes val | mIoU: 73.48 -> 74.03 | Improved |
| reuse gated local features three times (ablation) | Res-50+GCM+LCM(1) · Cityscapes val | mIoU: 74.03 -> 74.67 | Improved |
| cascade three adaptive context blocks instead of one (ablation) | Res-50+ACB · Cityscapes val | mIoU: 74.67 -> 76.53 | Improved |
| use ResNet-101 with three blocks instead of ResNet-50 (ablation) | ACB#3 · Cityscapes val | mIoU: 76.53 -> 77.42 | Improved |
| add multi-grid dilated convolution (ablation) | Res-101+ACB#3 · Cityscapes val | mIoU: 77.42 -> 78.50 | Improved |
| add multi-scale data augmentation (ablation) | Res-101+ACB#3+MG · Cityscapes val | mIoU: 78.50 -> 80.09 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- DANet — DANet: Divergent Activation for Weakly Supervised Object Localization (ICCV 2019) · better on 3 of 3 shared results
- CCNet — CCNet: Criss-Cross Attention for Semantic Segmentation (ICCV 2019) · better on 1 of 1 shared result
- BiSeNet — BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation (ECCV 2018) · better on 1 of 1 shared result
- 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
- Scene Parsing With Global Context Embedding (ICCV 2017 · similarity 0.90)
- Perspective-Adaptive Convolutions for Scene Parsing (TPAMI 2019 · similarity 0.90)
- Scale-Adaptive Convolutions for Scene Parsing (ICCV 2017 · similarity 0.90)
- Pyramid Scene Parsing Network (CVPR 2017 · similarity 0.90)
- GINet: Graph Interaction Network for Scene Parsing (ECCV 2020 · similarity 0.88)
- Strip Pooling: Rethinking Spatial Pooling for Scene Parsing (CVPR 2020 · similarity 0.88)
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.
