Research map / Detection and segmentation / CNN features
OSKDet: Orientation-Sensitive Keypoint Localization for Rotated Object Detection
Dongchen Lu, Dongmei Li, Yali Li, Shengjin Wang
Check 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 |
|---|---|---|---|
| Use the full OSKDet method with multi-scale training and testing | OSKDet · DOTA-v1.0 OBB | AP.5: 76.81 -> 77.81 | Improved |
| Add orientation-sensitive heatmap instead of standard Gaussian heatmap (ablation) | OSKDet · DOTA format comparison with FPN | AP.5: 75.54 -> 76.16 | Improved |
| Add rotation-aware deformable convolution (ablation) | OSKDet · DOTA validation | AP.5-.95: 42.86 -> 44.13 | Improved |
| Add standard deformable convolution (ablation) | OSKDet · DOTA validation | AP.5-.95: 42.86 -> 43.38 | Improved |
| Use 44° rather than unsorted keypoints (ablation) | OSKDet · DOTA keypoint sorting | AP.5: 75.11 -> 76.10 | Improved |
| Use 0° as the keypoint sorting interface (ablation) | OSKDet · DOTA keypoint sorting | AP.5: 75.11 -> 75.49 | Improved |
| Add multi-path feature fusion to 44° keypoint sorting (ablation) | OSKDet · DOTA keypoint sorting | AP.5: 76.10 -> 76.75 | Improved |
| Use modulated loss (ablation) | OSKDet · DOTA keypoint sorting | AP.5: 75.11 -> 75.96 | Improved |
| Add multi-scale training and testing (ablation) | OSKDet · DOTA ablation | AP.5: 73.38 -> 75.54 | Improved |
| Omit multi-scale training and testing from the full configuration (ablation) | OSKDet · DOTA ablation | AP.5: 77.81 -> 76.06 | Hurt |
| Add orientation-sensitive heatmap and rotation-aware convolution to multi-scale training (ablation) | OSKDet · DOTA ablation | AP.5: 75.54 -> 76.86 | Improved |
| Use OSKDet on ICDAR2015 instead of the strongest reported F-measure baseline | OSKDet · ICDAR2015 | F-measure: 92.10 -> 91.34 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SCRDet++ — SCRDet: Towards More Robust Detection for Small, Cluttered and Rotated Objects (ICCV 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PolarNet: Learning to Optimize Polar Keypoints for Keypoint Based Object Detection (ICLR 2021 · similarity 0.86)
- Learning High-Precision Bounding Box for Rotated Object Detection via Kullback-Leibler Divergence (NeurIPS 2021 · similarity 0.86)
- R3Det: Refined Single-Stage Detector with Feature Refinement for Rotating Object (AAAI 2021 · similarity 0.86)
- Adaptive Rotated Convolution for Rotated Object Detection (ICCV 2023 · similarity 0.86)
- Key-Grid: Unsupervised 3D Keypoints Detection using Grid Heatmap Features (NeurIPS 2024 · similarity 0.84)
- FRED: Towards a Full Rotation-Equivariance in Aerial Image Object Detection (AAAI 2024 · similarity 0.84)
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.
