Research map / Detection and segmentation / CNN features
Dynamic Multi-Scale Filters for Semantic Segmentation
Junjun He, Zhongying Deng, Yu Qiao
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 |
|---|---|---|---|
| use context-aware dynamic filters instead of fixed regular-conv filters (ablation) | DCM · PASCAL VOC 2012 val, kernel sizes {1} | mIoU (%): 70.76 -> 77.75 | Improved |
| use context-aware dynamic filters instead of fixed regular-conv filters (ablation) | DCM · PASCAL VOC 2012 val, kernel sizes {1,3,5,7} | mIoU (%): 72.37 -> 78.21 | Improved |
| use context-aware filters instead of fixed depthwise filters (ablation) | DCM · PASCAL VOC 2012 val, kernel sizes {1,3} | mIoU (%): 71.12 -> 78.52 | Improved |
| use context-aware filters instead of dilated filters {6,12,24,36} (ablation) | DCM · PASCAL VOC 2012 val | mIoU (%): 77.60 -> 78.21 | Improved |
| use context-aware filters instead of dilated filters {1,2,3,4} (ablation) | DCM · PASCAL VOC 2012 val | mIoU (%): 73.37 -> 78.21 | Improved |
| add more DCM branches (kernel sets {1} to {1,3,5}) (ablation) | DMNet · PASCAL VOC 2012 val | mIoU (%): 77.75 -> 78.76 | Improved |
| add a fourth DCM branch (kernel 7) to {1,3,5} (ablation) | DMNet · PASCAL VOC 2012 val | mIoU (%): 78.76 -> 78.21 | Hurt |
| generate filters from Stage5 instead of Stage2 context (ablation) | DMNet · PASCAL VOC 2012 val | mIoU (%): 73.13 -> 78.21 | Improved |
| add DCMs to FCN (ResNet50) (ablation) | FCN · PASCAL VOC 2012 val | mIoU (%): 69.83 -> 78.21 | Improved |
| add DCMs to FCN (ResNet101) (ablation) | FCN · PASCAL VOC 2012 val | mIoU (%): 76.24 -> 80.82 | Improved |
| use ResNet101 instead of ResNet50 | DMNet · PASCAL VOC 2012 val | mIoU (%): 78.21 -> 80.82 | Improved |
| add deep supervision (ablation) | DMNet ResNet101 · PASCAL VOC 2012 val | mIoU (%): 80.87 -> 81.08 | 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.
- DANet — DANet: Divergent Activation for Weakly Supervised Object Localization (ICCV 2019) · 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
- Attention to Scale: Scale-Aware Semantic Image Segmentation (CVPR 2016 · similarity 0.89)
- Multi-Scale Representations by Varying Window Attention for Semantic Segmentation (ICLR 2024 · similarity 0.88)
- Multi-Scale Context Intertwining for Semantic Segmentation (ECCV 2018 · similarity 0.88)
- Dynamic Sampling Network for Semantic Segmentation (AAAI 2020 · similarity 0.88)
- Hierarchically Gated Deep Networks for Semantic Segmentation (CVPR 2016 · similarity 0.88)
- Context Encoding for Semantic Segmentation (CVPR 2018 · 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.
