Research map / Detection and segmentation / CNN features
Scale-Adaptive Convolutions for Scene Parsing
Rui Zhang, Sheng Tang, Yongdong Zhang, Jintao Li, Shuicheng 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 |
|---|---|---|---|
| add scale-adaptive convolution in last layer, jointly trained (ablation) | Baseline ResNet-101 · Cityscapes val | mean IoU: 74.0 -> 75.9 | Improved |
| add scale-adaptive convolutions in last layer and 5th block 3x3 layers (ablation) | Baseline ResNet-101 · Cityscapes val | mean IoU: 74.0 -> 76.5 | Improved |
| learn only scale regression layer with other weights fixed (single) (ablation) | Baseline ResNet-101 · Cityscapes val | mean IoU: 74.0 -> 74.5 | Improved |
| learn only scale regression layers with weights fixed (multiple) (ablation) | Baseline ResNet-101 · Cityscapes val | mean IoU: 74.0 -> 74.8 | Improved |
| jointly train convolution parameters with scale coefficients instead of fixing them (ablation) | SAC-single · Cityscapes val | mean IoU: 74.5 -> 75.9 | Improved |
| use multiple scale-adaptive layers instead of single (ablation) | SAC · Cityscapes val | mean IoU: 75.9 -> 76.5 | Improved |
| add multi-scale fusion at test time (ablation) | SAC-multiple · Cityscapes val | mean IoU: 76.5 -> 78.7 | Improved |
| add scale-adaptive convolution in last layer, jointly trained (ablation) | Baseline ResNet-101 · ADE20K val | mean IoU: 39.75 -> 41.68 | Improved |
| add multiple scale-adaptive layers (ablation) | Baseline ResNet-101 · ADE20K val | mean IoU: 39.75 -> 42.38 | Improved |
| fixed weights, only scale regression trained (single) (ablation) | Baseline ResNet-101 · ADE20K val | mean IoU: 39.75 -> 40.32 | Improved |
| fixed weights, only scale regression trained (multiple) (ablation) | Baseline ResNet-101 · ADE20K val | mean IoU: 39.75 -> 40.87 | Improved |
| add multi-scale fusion at test time (ablation) | SAC-multiple · ADE20K val | mean IoU: 42.38 -> 44.30 | Improved |
Closest papers by meaning
- Perspective-Adaptive Convolutions for Scene Parsing (TPAMI 2019 · similarity 0.93)
- Adaptive Context Network for Scene Parsing (ICCV 2019 · similarity 0.90)
- Learning Adaptive Receptive Fields for Deep Image Parsing Network (CVPR 2017 · similarity 0.88)
- Scene Recognition With CNNs: Objects, Scales and Dataset Bias (CVPR 2016 · similarity 0.87)
- Malleable 2.5D Convolution: Learning Receptive Fields along the Depth-axis for RGB-D Scene Parsing (ECCV 2020 · similarity 0.86)
- Video Scene Parsing With Predictive Feature Learning (ICCV 2017 · similarity 0.86)
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.
