Research map / Architectures and efficiency / CNN design
Dynamic Region-Aware Convolution
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace 1×1 convolutions with 8-region DRConv | ShuffleNetV2 0.5× · ImageNet 2012 val | Top-1 accuracy (%): 60.8 -> 67.1 | Helped |
| Replace 1×1 convolutions with 8-region DRConv | ShuffleNetV2 1× · ImageNet 2012 val | Top-1 accuracy (%): 69.5 -> 73.1 | Helped |
| Replace 1×1 convolutions with 8-region DRConv | MobileNetV2 · ImageNet 2012 val | Top-1 accuracy (%): 72.0 -> 75.7 | Helped |
| Replace 1×1 convolutions with 8-region DRConv | MobileNetV1 · ImageNet 2012 val | Top-1 accuracy (%): 70.6 -> 75.5 | Helped |
| Replace bottleneck 1×1 convolutions with DRConv | MobileFaceNet · MegaFace rank-1 identification, 1M distractors | Accuracy (%): 91.3 -> 96.2 | Helped |
| Replace backbone 1×1 convolutions with 8-region DRConv | DetNAS-300M · COCO val detection | APbbox: 36.6 -> 38.4 | Helped |
| Add 4-region DRConv to two FPN layers (ablation) | Mask R-CNN · COCO val detection | APbbox: 39.1 -> 39.8 | Helped |
| Increase DRConv regions from 4 to 8 (ablation) | Mask R-CNN FPN · COCO val detection | APbbox: 39.8 -> 40.2 | Helped |
| Increase DRConv regions from 8 to 16 (ablation) | Mask R-CNN FPN · COCO val detection | APbbox: 40.2 -> 40.3 | Helped |
| Add 4-region DRConv to two FPN layers (ablation) | Mask R-CNN · COCO val segmentation | APmask: 34.5 -> 35.3 | Helped |
| Increase DRConv regions from 4 to 8 (ablation) | Mask R-CNN FPN · COCO val segmentation | APmask: 35.3 -> 35.5 | Helped |
| Increase DRConv regions from 8 to 16 (ablation) | Mask R-CNN FPN · COCO val segmentation | APmask: 35.5 -> 35.6 | Helped |
Closest papers by meaning
- Omni-Dimensional Dynamic Convolution (ICLR 2022 · similarity 0.87)
- TVConv: Efficient Translation Variant Convolution for Layout-Aware Visual Processing (CVPR 2022 · similarity 0.87)
- Dynamic Convolution: Attention Over Convolution Kernels (CVPR 2020 · similarity 0.87)
- Glance and Focus: a Dynamic Approach to Reducing Spatial Redundancy in Image Classification (NeurIPS 2020 · similarity 0.86)
- Frequency Dynamic Convolution for Dense Image Prediction (CVPR 2025 · similarity 0.86)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · 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.