Research map / Detection and segmentation / CNN features
Arbitrary-Oriented Object Detection with Circular Smooth Label
Xue Yang, Junchi Yan
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 |
|---|---|---|---|
| replace angle regression with circular smooth label classification (Gaussian, 180) | RetinaNet-H · DOTA | 5-mAP (%): 51.00 -> 56.21 (+5.21) | Improved |
| replace angle regression with CSL (Gaussian, 180) | RetinaNet-H · DOTA | mAP (%): 64.10 -> 64.50 (+0.40) | Improved |
| replace angle regression with CSL (Gaussian, 90) | RetinaNet-R · DOTA | mAP (%): 62.76 -> 65.45 (+2.69) | Improved |
| replace angle regression with CSL (Gaussian, 180) | FPN-H · DOTA | mAP (%): 69.45 -> 70.92 (+1.47) | Improved |
| replace angle regression with CSL (Gaussian, 90) | FPN-H · DOTA | 5-mAP (%): 61.85 -> 63.71 (+1.86) | Improved |
| use Gaussian window instead of pulse window (180) (ablation) | CSL RetinaNet-H · DOTA | mAP (%): 42.06 -> 64.50 | Improved |
| use Gaussian window instead of triangle window (180) (ablation) | CSL RetinaNet-H · DOTA | mAP (%): 57.94 -> 64.50 | Improved |
| use Gaussian window instead of rectangular window (180) (ablation) | CSL RetinaNet-H · DOTA | mAP (%): 61.98 -> 64.50 | Improved |
| use 180 angle range instead of 90 with Gaussian window (avoids EoE) (ablation) | CSL RetinaNet-H · DOTA | mAP (%): 63.51 -> 64.50 | Improved |
| increase window radius from 0 to 6 (ablation) | CSL RetinaNet-H · DOTA | mAP (%): 40.78 -> 64.50 | Improved |
| increase window radius from 0 to 6 (ablation) | CSL FPN-H · DOTA | mAP (%): 48.08 -> 70.92 | Improved |
| increase window radius from 6 to 8 (ablation) | CSL RetinaNet-H · DOTA | mAP (%): 64.50 -> 63.99 | Hurt |
Closest papers by meaning
- On the Arbitrary-Oriented Object Detection: Classification Based Approaches Revisited (IJCV 2022 · similarity 0.95)
- Dense Label Encoding for Boundary Discontinuity Free Rotation Detection (CVPR 2021 · similarity 0.92)
- On Boundary Discontinuity in Angle Regression Based Arbitrary Oriented Object Detection (TPAMI 2024 · similarity 0.89)
- Towards Rotation Invariance in Object Detection (ICCV 2021 · similarity 0.89)
- Rethinking Boundary Discontinuity Problem for Oriented Object Detection (CVPR 2024 · similarity 0.89)
- FRED: Towards a Full Rotation-Equivariance in Aerial Image Object Detection (AAAI 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.
