Research map / Detection and segmentation / Weakly supervised detection and segmentation
Point2RBox-v3: Self-Bootstrapping from Point Annotations via Integrated Pseudo-Label Refinement and Utilization
Teng Zhang, Ziqian Fan, Mingxin Liu, Xin Zhang, Xudong Lu, Wentong Li, Yue Zhou, Yi Yu, Xiang Li, Junchi Yan, Xue Yang
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 |
|---|---|---|---|
| add Progressive Label Assignment (multi-level FPN assignment from pseudo labels) (ablation) | Point2RBox-v2 · DOTA-v1.0 end-to-end | AP50: 51.0 -> 56.6 | Improved |
| add PLA (ablation) | Point2RBox-v2 · DOTA-v1.0 two-stage | AP50: 62.6 -> 64.6 | Improved |
| add Prior-Guided Dynamic Mask Loss (SAM + watershed routing) (ablation) | Point2RBox-v2 · DOTA-v1.0 end-to-end | AP50: 51.0 -> 54.2 | Improved |
| combine PLA and PGDM-Loss (ablation) | Point2RBox-v2 · DOTA-v1.0 end-to-end | AP50: 51.0 -> 59.6 | Improved |
| switch PLA from watershed boxes to network predictions mid-training (epoch 3-6) instead of at epoch 0 (ablation) | PLA · DOTA-v1.0 end-to-end | AP50: 56.6 -> 59.6 | Improved |
| use only watershed boxes for the whole training (switch epoch 12) (ablation) | PLA · DOTA-v1.0 end-to-end | AP50: 59.6 -> 56.3 | Hurt |
| route all instances to SAM (N_thr infinity) instead of only sparse scenes (ablation) | PGDM-Loss · DOTA-v1.0 end-to-end | AP50: 59.6 -> 57.2 | Hurt |
| increase sparse-scene threshold N_thr from 4 to 8 (ablation) | PGDM-Loss · DOTA-v1.0 end-to-end | AP50: 59.6 -> 58.9 | No clear effect |
| use prior-guided mask selection instead of SAM native confidence (ablation) | PGDM-Loss · DOTA-v1.0 end-to-end | AP50: 57.86 -> 59.61 | Improved |
| use prior-guided mask selection instead of SAM native confidence (ablation) | PGDM-Loss · DOTA-v1.0 two-stage | AP50: 63.59 -> 66.09 | Improved |
| add PLA and PGDM to PWOOD with 10% point labels | PWOOD · DOTA-v1.0 | AP50: 42.35 -> 50.67 | Improved |
| add PLA and PGDM to PWOOD with 10% point labels | PWOOD · DOTA-v1.5 | AP50: 35.33 -> 43.25 | Improved |
Closest papers by meaning
- Point2RBox-v2: Rethinking Point-supervised Oriented Object Detection with Spatial Layout Among Instances (CVPR 2025 · similarity 0.90)
- Point2RBox: Combine Knowledge from Synthetic Visual Patterns for End-to-end Oriented Object Detection with Single Point Supervision (CVPR 2024 · similarity 0.89)
- Point-Teaching: Weakly Semi-supervised Object Detection with Point Annotations (AAAI 2023 · similarity 0.89)
- PointOBB-v2: Towards Simpler, Faster, and Stronger Single Point Supervised Oriented Object Detection (ICLR 2025 · similarity 0.87)
- Leveraging Imagery Data with Spatial Point Prior for Weakly Semi-supervised 3D Object Detection (AAAI 2024 · similarity 0.87)
- Fully Data-Driven Pseudo Label Estimation for Pointly-Supervised Panoptic Segmentation (AAAI 2024 · similarity 0.87)
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.
