Research map / Detection and segmentation / CNN features
Polygon-to-Polygon Distance Loss for Rotated Object Detection
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 smoothL1 with P2P Loss | RetinaNet on DOTA · DOTA | mAP: 69.916 -> 71.055 | Helped |
| Add center loss and semiperimeter loss to the P2P distance (ablation) | P2P Loss on DOTA · DOTA | mAP: 70.496 -> 70.911 | Helped |
| Add center loss while retaining P2P distance and semiperimeter loss (ablation) | P2P Loss on DOTA · DOTA | mAP: 70.717 -> 70.897 | Helped |
| Add semiperimeter loss while retaining P2P distance and center loss (ablation) | P2P Loss on DOTA · DOTA | mAP: 70.496 -> 70.911 | Helped |
| Add P2P Loss to RATSS and QFL (ablation) | RetinaNet on DOTA · DOTA | mAP: 71.421 -> 72.200 | Helped |
| Use RATSS together with QFL instead of either alone (ablation) | RetinaNet on DOTA · DOTA | mAP: 70.772; 71.055 -> 71.421 | Helped |
| Increase smoothL1 anchor count from 3 to 6 (ablation) | RetinaNet-FPN-R-50 on HRSC2016 · HRSC2016 | mAP: 73.512 -> 83.532 | Helped |
| Increase P2P Loss# anchor count from 3 to 6 (ablation) | RetinaNet-FPN-R-50 on HRSC2016 · HRSC2016 | mAP: 89.177 -> 88.675 | No clear effect |
| Use P2P Loss# rather than smoothL1 with 3 anchors (ablation) | RetinaNet-FPN-R-50 on HRSC2016 · HRSC2016 | mAP: 73.512 -> 89.177 | Helped |
| Switch smoothL1 from Doc to Dle box definition (ablation) | RetinaNet on DOTA · DOTA | mAP: 69.416 -> 69.916 | Helped |
| Change box definition and coder under P2P Loss# (ablation) | RetinaNet on DOTA · DOTA | mAP: 72.380; 72.211; 72.200 | No clear effect |
| Use P2P Loss# instead of the other single-stage detectors listed | RetinaNet on DOTA · DOTA test | mAP: 79.2; 80.1 | Mixed |
Closest papers by meaning
- Learning Modulated Loss for Rotated Object Detection (AAAI 2021 · similarity 0.85)
- The KFIoU Loss for Rotated Object Detection (ICLR 2023 · similarity 0.83)
- Theoretically Achieving Continuous Representation of Oriented Bounding Boxes (CVPR 2024 · similarity 0.81)
- PolarNeXt: Rethink Instance Segmentation with Polar Representation (CVPR 2025 · similarity 0.81)
- Rethinking Boundary Discontinuity Problem for Oriented Object Detection (CVPR 2024 · similarity 0.81)
- Point2RBox: Combine Knowledge from Synthetic Visual Patterns for End-to-end Oriented Object Detection with Single Point Supervision (CVPR 2024 · similarity 0.80)
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.