Research Article
Local-to-Global Point Supervised Object Detector via Aggregation of Discriminative Parts
@INPROCEEDINGS{10.1007/978-3-031-71716-1_12, author={Yidan Zhang and Yingyan Hou and Xiaoxuan Liu and Xiaohe Li and Fangli Mou and Peirong Zhang and Xiyu Qi and Jie Jia and Lei Wang and Xinyu Zhao}, title={Local-to-Global Point Supervised Object Detector via Aggregation of Discriminative Parts}, proceedings={Machine Learning and Intelligent Communication. 8th EAI International Conference, MLICOM 2023, Beijing, China, December 17, 2023, Proceedings}, proceedings_a={MLICOM}, year={2024}, month={9}, keywords={Object Detection Point Supervised Object Detection}, doi={10.1007/978-3-031-71716-1_12} }- Yidan Zhang
Yingyan Hou
Xiaoxuan Liu
Xiaohe Li
Fangli Mou
Peirong Zhang
Xiyu Qi
Jie Jia
Lei Wang
Xinyu Zhao
Year: 2024
Local-to-Global Point Supervised Object Detector via Aggregation of Discriminative Parts
MLICOM
Springer
DOI: 10.1007/978-3-031-71716-1_12
Abstract
Advanced fully supervised detectors benefit from abundant bounding-box annotations which accurately cover multi-scale objects. However, for point supervised object detectors (PSOD), each object is annotated by a single point without the information of scales. Some scholars have started to represent the scale of the objects through point-to-box regression, yet the accuracy is constrained by manual heuristic algorithms or local part activations. In this paper, we propose a Local-to-global Point Supervised Object Detector called LPSNet, which can adaptively generate globally aware pseudo bounding boxes. Initially, Point-level Prediction (PLP) precisely identifies the object’s location at a point level. Subsequently, Box-level Prediction with Aggregation of Discriminative Parts (BPAP) dynamically performs regression from points to component-level proposals, and consolidate those part proposals into global ones. Finally, under the supervision of the pseudo proposals, LPSNet with region proposal network and detection head attached obtains detection results. Extensive experiments on the MSCOCO datasets underlines the superiority of our approach, outperforming other available PSOD methodologies.
