Research map / Detection and segmentation / CNN features
RankDetNet: Delving Into Ranking Constraints for Object Detection
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 |
|---|---|---|---|
| Add global, class-specific, and IoU-guided ranking losses | RetinaNet with ResNet-50 · COCO 2017 test-dev | AP: 35.7 -> 38.2 | Helped |
| Add global, class-specific, and IoU-guided ranking losses | FCOS-ATSS with ResNet-50-DCN · COCO 2017 test-dev | AP: 43.0 -> 45.4 | Helped |
| Replace SA-SSD focal classification loss with ranking losses | SA-SSD · KITTI moderate car | AP@(IoU=0.7): +2.02% | Helped |
| Use global ranking loss only rather than the full set of ranking constraints (ablation) | RankDetNet with RetinaNet (ResNet-50) · COCO 2017 validation | AP: 34.2 vs. 37.8 | Hurt |
| Use class-specific ranking loss only rather than the full set of ranking constraints (ablation) | RankDetNet with RetinaNet (ResNet-50) · COCO 2017 validation | AP: 32.7 vs. 37.8 | Hurt |
| Add IoU-guided ranking loss to global and class-specific ranking losses (ablation) | RankDetNet with RetinaNet (ResNet-50) · COCO 2017 validation | AP: 36.6 -> 37.8 | Helped |
| Add dynamic reweighting factors to the global and class-specific losses (ablation) | RankDetNet with RetinaNet (ResNet-50) · COCO 2017 validation | AP: 36.6 -> 37.8 | Helped |
| Compute IoU-guided pairs separately per pyramid level instead of merging levels (ablation) | RankDetNet with RetinaNet (ResNet-50) · COCO 2017 validation | AP: 36.7 -> 37.8 | Helped |
| Stop gradients through the paired IoU score with lower object confidence (ablation) | IoU-guided ranking loss in RankDetNet · COCO 2017 validation | AP: 35.4 -> 37.8 | Helped |
| Use grouped negative mining instead of OHEM (ablation) | RankDetNet with RetinaNet (ResNet-50) · COCO 2017 validation | AP: 37.3 -> 37.8 | Helped |
| Replace the exponential ranking function with sigmoid or logistic (ablation) | RankDetNet with RetinaNet (ResNet-50) · COCO 2017 validation | AP: 37.3; 37.2; 37.3 | No clear effect |
| Set IoU-guided loss α to 5.0 rather than the selected α=1.0 (ablation) | RankDetNet with RetinaNet (ResNet-50) · COCO 2017 validation | AP: 37.3 -> 36.0 | Hurt |
Closest papers by meaning
- DR Loss: Improving Object Detection by Distributional Ranking (CVPR 2020 · similarity 0.91)
- Rank-DETR for High Quality Object Detection (NeurIPS 2023 · similarity 0.90)
- Optimizing Rank-Based Metrics With Blackbox Differentiation (CVPR 2020 · similarity 0.88)
- Revisiting AP Loss for Dense Object Detection: Adaptive Ranking Pair Selection (CVPR 2022 · similarity 0.87)
- LocNet: Improving Localization Accuracy for Object Detection (CVPR 2016 · similarity 0.87)
- Improving Single Domain-Generalized Object Detection: A Focus on Diversification and Alignment (CVPR 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.