GeoTeacher: Geometry-Guided 3D Object Detection
Research Paper#3D Object Detection, Semi-Supervised Learning, Computer Vision🔬 Research|Analyzed: Jan 3, 2026 19:10•
Published: Dec 29, 2025 02:24
•1 min read
•ArXivAnalysis
This paper addresses the challenge of semi-supervised 3D object detection, focusing on improving the student model's understanding of object geometry, especially with limited labeled data. The core contribution lies in the GeoTeacher framework, which uses a keypoint-based geometric relation supervision module to transfer knowledge from a teacher model to the student, and a voxel-wise data augmentation strategy with a distance-decay mechanism. This approach aims to enhance the student's ability in object perception and localization, leading to improved performance on benchmark datasets.
Key Takeaways
- •Proposes GeoTeacher, a novel framework for semi-supervised 3D object detection.
- •Introduces a keypoint-based geometric relation supervision module to transfer knowledge.
- •Employs a voxel-wise data augmentation strategy with a distance-decay mechanism.
- •Achieves state-of-the-art results on ONCE and Waymo datasets.
Reference / Citation
View Original"GeoTeacher enhances the student model's ability to capture geometric relations of objects with limited training data, especially unlabeled data."