Research map / Detection and segmentation / CNN features
Class-wise Dynamic Graph Convolution for Semantic Segmentation
Hanzhe Hu, Deyi Ji, Weihao Gan, Shuai Bai, Wei Wu, Junjie Yan
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 |
|---|---|---|---|
| perform graph reasoning class-wise instead of over all pixels (plain GCN) (ablation) | ResNet-101 + GCN · Cityscapes val | mIoU (%): 78.2 -> 79.4 | Improved |
| add plain GCN to the ResNet-101 baseline (ablation) | ResNet-101 · Cityscapes val | mIoU (%): 76.3 -> 78.2 | Improved |
| add CDGC module to the ResNet-101 baseline (ablation) | ResNet-101 · Cityscapes val | mIoU (%): 76.3 -> 79.4 | Improved |
| add ASPP instead of CDGC to ResNet-101 (ablation) | ResNet-101 · Cityscapes val | mIoU (%): 76.3 -> 78.4 | Improved |
| add CDGC on top of ResNet-101 + ASPP (concat) (ablation) | ResNet-101 + ASPP · Cityscapes val | mIoU (%): 78.4 -> 80.0 | Improved |
| fuse refined feature by concatenation instead of summation (ablation) | ResNet-101 + CDGC · Cityscapes val | mIoU (%): 79.2 -> 79.4 | Improved |
| fuse refined feature by concatenation instead of summation (ablation) | ResNet-101 + ASPP + CDGC · Cityscapes val | mIoU (%): 79.9 -> 80.0 | Improved |
| replace similarity graph with dynamic sampling at easy-positive ratio 1.0 (ablation) | ResNet-101 + ASPP + CDGC · Cityscapes val | mIoU (%): 80.0 -> 81.1 | Improved |
| dynamic sampling with a low easy positive ratio of 0.2 vs similarity graph (ablation) | ResNet-101 + ASPP + CDGC · Cityscapes val | mIoU (%): 80.0 -> 79.8 | Hurt |
| increase easy positive sampling ratio from 0.2 to 1.0 (ablation) | dynamic sampling CDGC · Cityscapes val | mIoU (%): 79.8 -> 81.1 | Improved |
| add hard positive samples to easy positive (ablation) | CDGC dynamic sampling · Cityscapes val | mIoU (%): 79.9 -> 80.5 | Improved |
| add hard negative samples to easy positive (ablation) | CDGC dynamic sampling · Cityscapes val | mIoU (%): 79.9 -> 80.0 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DANet — DANet: Divergent Activation for Weakly Supervised Object Localization (ICCV 2019) · better on 3 of 3 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
- Coupled Segmentation and Edge Learning via Dynamic Graph Propagation (NeurIPS 2021 · similarity 0.89)
- Dynamic Graph Message Passing Networks for Visual Recognition (TPAMI 2022 · similarity 0.88)
- Spatial Pyramid Based Graph Reasoning for Semantic Segmentation (CVPR 2020 · similarity 0.88)
- Lightweight and Progressively-Scalable Networks for Semantic Segmentation (IJCV 2023 · similarity 0.87)
- Learning to Predict Context-adaptive Convolution for Semantic Segmentation (ECCV 2020 · similarity 0.87)
- Learning Dynamic Routing for Semantic Segmentation (CVPR 2020 · 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.
