PointOBB-v3: Expanding Performance Boundaries of Single Point-Supervised Oriented Object Detection

2025-01-23

AI Summary

This paper addresses end-to-end rotated object detection under single-point supervision—without requiring additional priors to generate high-quality pseudo-rotated bounding boxes. Methodologically, it introduces tri-view collaborative augmentation (original, scaled, and rotation-flipped views), a scale-sensitive consistency loss (SSC), a scale-sensitive feature fusion module (SSFF), and geometry-symmetry-driven self-supervised angle learning. It further achieves the first end-to-end out-of-distribution (OOD) detection framework under single-point supervision and proposes an instance-aware weighting (IAW) strategy to enhance hard-example learning. Evaluated on six remote sensing benchmarks—DIOR-R, DOTA-v1.0/v1.5/v2.0, FAIR1M, STAR, and RSAR—the method achieves a 3.56% average precision gain over state-of-the-art approaches, demonstrating substantial improvements in both detection accuracy and generalization.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Multi-instance/Multi-view Learning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Security and Privacy: Data transparency and provenance

Abstract

With the growing demand for oriented object detection (OOD), recent studies on point-supervised OOD have attracted significant interest. In this paper, we propose PointOBB-v3, a stronger single point-supervised OOD framework. Compared to existing methods, it generates pseudo rotated boxes without additional priors and incorporates support for the end-to-end paradigm. PointOBB-v3 functions by integrating three unique image views: the original view, a resized view, and a rotated/flipped (rot/flp) view. Based on the views, a scale augmentation module and an angle acquisition module are constructed. In the first module, a Scale-Sensitive Consistency (SSC) loss and a Scale-Sensitive Feature Fusion (SSFF) module are introduced to improve the model's ability to estimate object scale. To achieve precise angle predictions, the second module employs symmetry-based self-supervised learning. Additionally, we introduce an end-to-end version that eliminates the pseudo-label generation process by integrating a detector branch and introduces an Instance-Aware Weighting (IAW) strategy to focus on high-quality predictions. We conducted extensive experiments on the DIOR-R, DOTA-v1.0/v1.5/v2.0, FAIR1M, STAR, and RSAR datasets. Across all these datasets, our method achieves an average improvement in accuracy of 3.56% in comparison to previous state-of-the-art methods. The code will be available at https://github.com/ZpyWHU/PointOBB-v3.

Problem

• Rotational Object Detection
• Accuracy Improvement
• Efficiency Enhancement

Innovation

• Rotation-invariant object detection
• Scale-aware module
• Direction regression module