Research map / Detection and segmentation / CNN features
Object Detection Based on Region Decomposition and Assembly
Read the paperCheck 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 |
|---|---|---|---|
| Add region decomposition and assembly to Faster R-CNN (ablation) | Faster R-CNN on PASCAL VOC07 · PASCAL VOC07 test | mAP: 68.90 -> 71.95 | Helped |
| Add multi-scale proposals using five scales (ablation) | R-DAD on PASCAL VOC07 · PASCAL VOC07 test | mAP: 71.95 -> 73.90 | Helped |
| Add up-sampling of the feature map (ablation) | Five-scale R-DAD on PASCAL VOC07 · PASCAL VOC07 test | mAP: 73.90 -> 74.90 | Helped |
| Use five proposal scales instead of three (ablation) | Multi-scale R-DAD on PASCAL VOC07 · PASCAL VOC07 test | mAP: 72.65 -> 73.90 | Helped |
| Use Max aggregation at all stages rather than Sum aggregation (ablation) | RDA network on PASCAL VOC07 · PASCAL VOC07 test | mAP: 69.61 -> 74.90 | Helped |
| Use concatenation instead of Max aggregation at stage 3 (ablation) | RDA network on PASCAL VOC07 · PASCAL VOC07 test | mAP: 74.90 -> 71.95 | Hurt |
| Use dilated convolutions or 5x5 filters instead of 3x3 filters (ablation) | RDA network on PASCAL VOC07 · PASCAL VOC07 test | mAP: 74.90; 75.10 | No clear effect |
| Use R-DAD instead of Faster R-CNN with ResNet101 | Object detection on PASCAL VOC07 · PASCAL VOC07 test | mAP: 74.9 -> 77.6 | Helped |
| Use R-DAD instead of Faster R-CNN with ResNet101 | Object detection on PASCAL VOC07++12 · PASCAL VOC07 test | mAP: 76.6 -> 81.2 | Helped |
| Use R-DAD with ResNet152 rather than ResNet101 | Object detection on PASCAL VOC2012 · PASCAL VOC2012 test | mAP: 80.2 -> 82.0 | Helped |
| Apply multi-scale testing | R-DAD with ResNet-101 on MSCOCO18 · MSCOCO18 test-dev | mAP (IoU 0.5:0.95): 40.4 -> 43.1 | Helped |
Closest papers by meaning
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.91)
- CRAFT Objects From Images (CVPR 2016 · similarity 0.89)
- R-FCN++: Towards Accurate Region-Based Fully Convolutional Networks for Object Detection (AAAI 2018 · similarity 0.87)
- Recognize Any Regions (NeurIPS 2024 · similarity 0.86)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.86)
- Generative Region-Language Pretraining for Open-Ended Object Detection (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.