Research map / Detection and segmentation / CNN features
Cascade RPN: Delving into High-Quality Region Proposal Network with Adaptive Convolution
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 |
|---|---|---|---|
| Replace standard RPN with Cascade RPN | RPN · COCO 2017 val | AR1000: 58.3 -> 71.7 | Helped |
| Replace RPN proposals with Cascade RPN proposals | Fast R-CNN · COCO 2017 test-dev, 300 proposals | AP: 36.6 -> 40.1 | Helped |
| Replace RPN proposals with Cascade RPN proposals | Faster R-CNN · COCO 2017 test-dev, 300 proposals | AP: 36.9 -> 40.6 | Helped |
| Reduce anchors from 3 per location to 1 (ablation) | RPN · COCO 2017 val | AR1000: 58.3 -> 55.8 | Hurt |
| Add multi-stage refinement without alignment (ablation) | Single-anchor RPN · COCO 2017 val | AR1000: 55.8 -> 58.0 | Helped |
| Add adaptive convolution alignment (ablation) | Single-anchor cascade · COCO 2017 val | AR1000: 58.0 -> 67.8 | Helped |
| Add anchor-free stage 1 and anchor-based stage 2 sample metrics (ablation) | Aligned Cascade RPN · COCO 2017 val | AR1000: 67.8 -> 68.6 | Helped |
| Add regression statistics (ablation) | Cascade RPN with AFAB metrics · COCO 2017 val | AR1000: 68.6 -> 71.5 | Helped |
| Add IoU loss (ablation) | Cascade RPN with regression statistics · COCO 2017 val | AR1000: 71.5 -> 71.7 | Helped |
| Add center alignment (ablation) | Cascade RPN · COCO 2017 val | AR1000: 58.0 -> 64.1 | Helped |
| Add shape alignment to center alignment (ablation) | Cascade RPN · COCO 2017 val | AR1000: 64.1 -> 67.8 | Helped |
| Use anchor-free metrics at every stage instead of progressive AF then AB metrics (ablation) | Cascade RPN · COCO 2017 val | AR1000: 68.6 -> 66.4 | Hurt |
Closest papers by meaning
- Siamese Cascaded Region Proposal Networks for Real-Time Visual Tracking (CVPR 2019 · similarity 0.89)
- FA-RPN: Floating Region Proposals for Face Detection (CVPR 2019 · similarity 0.86)
- Joint Training of Cascaded CNN for Face Detection (CVPR 2016 · similarity 0.86)
- Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 2018 · similarity 0.86)
- CRAFT Objects From Images (CVPR 2016 · similarity 0.86)
- R-FCN++: Towards Accurate Region-Based Fully Convolutional Networks for Object Detection (AAAI 2018 · 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.