Research map / Detection and segmentation / Weakly supervised detection and segmentation
H2RBox-v2: Incorporating Symmetry for Boosting Horizontal Box Supervised Oriented 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 |
|---|---|---|---|
| add PSC angle coder and snap loss together in SS branch (ablation) | H2RBox-v2 · DOTA-v1.0 | AP50: 52.24 -> 72.31 | Helped |
| add snap loss on top of PSC (ablation) | H2RBox-v2 · DOTA-v1.0 | AP50: 27.57 -> 72.31 | Helped |
| add PSC on top of snap loss (ablation) | H2RBox-v2 · HRSC | AP50: 88.52 -> 89.66 | Helped |
| add CircumIoU loss (no rotation aug) (ablation) | H2RBox-v2 · DOTA-v1.0 | AP50: 71.49 -> 72.31 | Helped |
| add random rotation augmentation with CircumIoU loss (ablation) | H2RBox-v2 · DOTA-v1.0 | AP50: 72.31 -> 71.79 | No clear effect |
| add random rotation augmentation with CircumIoU loss (ablation) | H2RBox-v2 · HRSC | AP50: 89.66 -> 89.56 | No clear effect |
| add random rotation augmentation to IoU-loss variant (ablation) | H2RBox-v2 without CircumIoU · DOTA-v1.0 | AP50: 71.49 -> 29.34 | Hurt |
| add flip consistency loss (lambda 0 -> 0.05) (ablation) | H2RBox-v2 · HRSC | AP50: 0.32 -> 89.66 | Helped |
| add flip consistency loss (lambda 0 -> 0.05) (ablation) | H2RBox-v2 · DOTA-v1.0 | AP50: 66.37 -> 72.31 | Helped |
| raise flip loss weight lambda from 0.05 to 1.0 (ablation) | H2RBox-v2 · HRSC | AP50: 89.66 -> 6.97 | Hurt |
| use reflection padding instead of zeros in rotated view (ablation) | H2RBox-v2 · HRSC | AP50: 89.32 -> 89.66 | Helped |
| use H2RBox-v2 instead of H2RBox under HBox noise sigma=30% (ablation) | H2RBox-v2 · DOTA-v1.0 | AP50: 67.39 -> 71.11 | Helped |
Closest papers by meaning
- Point2RBox: Combine Knowledge from Synthetic Visual Patterns for End-to-end Oriented Object Detection with Single Point Supervision (CVPR 2024 · similarity 0.90)
- ABBSPO: Adaptive Bounding Box Scaling and Symmetric Prior based Orientation Prediction for Detecting Aerial Image Objects (CVPR 2025 · similarity 0.89)
- BGHR: Bridging the Gap Between HBox-Supervised and RBox-Supervised Oriented Object Detection via Adaptive Fine-Grained Sample Mining (AAAI 2025 · similarity 0.88)
- R2Det: Exploring Relaxed Rotation Equivariance in 2D Object Detection (ICLR 2025 · similarity 0.86)
- PointOBB-v2: Towards Simpler, Faster, and Stronger Single Point Supervised Oriented Object Detection (ICLR 2025 · similarity 0.85)
- Relational Matching for Weakly Semi-Supervised Oriented Object Detection (CVPR 2024 · 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.