Research map / Detection and segmentation / CNN features
Deformable Convolutional Networks
Jifeng Dai, Haozhi Qi, Yuwen Xiong, Yi Li, Guodong Zhang, Han Hu, Yichen Wei
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 deformable convolution to the last 3 conv layers (ablation) | ResNet-101 DeepLab · PASCAL VOC | mIoU@V: 69.7 -> 75.2 | Improved |
| add deformable convolution to the last 3 conv layers (ablation) | class-aware RPN · VOC | mAP@0.7: 44.9 -> 57.2 | Improved |
| add deformable convolution to the last 3 conv layers (ablation) | Faster R-CNN · VOC | mAP@0.5: 78.1 -> 78.6 | Improved |
| add deformable convolution to the last 3 conv layers (ablation) | R-FCN · VOC | mAP@0.7: 61.8 -> 64.7 | Improved |
| increase deformable layers from 3 to 6 (ablation) | DeepLab · PASCAL VOC | mIoU@V: 75.2 -> 74.8 | No clear effect |
| increase deformable layers from 3 to 6 (ablation) | R-FCN · VOC | mAP@0.7: 64.7 -> 65.4 | Improved |
| replace atrous convolution (dilation 2) by deformable convolution (ablation) | DeepLab · PASCAL VOC | mIoU@V: 73.6 -> 75.3 | Improved |
| use larger atrous dilation (2 -> 4) (ablation) | Faster R-CNN · VOC | mAP@0.5: 78.1 -> 78.6 | Improved |
| use largest atrous dilation (8 vs 4) (ablation) | Faster R-CNN · VOC | mAP@0.5: 78.6 -> 77.8 | Hurt |
| add deformable RoI pooling alone (ablation) | R-FCN · VOC | mAP@0.7: 61.8 -> 65.0 | Improved |
| add deformable RoI pooling alone (ablation) | Faster R-CNN · VOC | mAP@0.7: 62.1 -> 66.6 | Improved |
| combine deformable convolution and RoI pooling (ablation) | R-FCN · VOC | mAP@0.7: 64.7 -> 68.5 | Improved |
Closest papers by meaning
- Deformable ConvNets V2: More Deformable, Better Results (CVPR 2019 · similarity 0.92)
- Deformable Kernels: Adapting Effective Receptive Fields for Object Deformation (ICLR 2020 · similarity 0.88)
- DeepID-Net: Object Detection with Deformable Part Based Convolutional Neural Networks (TPAMI 2016 · similarity 0.88)
- Deformable Kernel Networks for Joint Image Filtering (IJCV 2020 · similarity 0.88)
- Deformable Graph Convolutional Networks (AAAI 2022 · similarity 0.88)
- Efficient Deformable ConvNets: Rethinking Dynamic and Sparse Operator for Vision Applications (CVPR 2024 · similarity 0.85)
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.
