V2V4Real, OPV2V
收藏资源简介:
本研究介绍了一种名为DOtA的无监督3D目标检测方法,该方法利用多智能体LiDAR扫描的内部共享信息进行训练,无需外部标签。DOtA包括初步标签生成、多尺度边界框编码标签过滤和标签内部对比学习三个部分。该方法在V2V4Real和OPV2V数据集上进行了测试,证明了其优越性。数据集的具体信息未在文中详细描述,但可以从提供的数据集中了解到,这些数据集涉及多智能体协同观测,并用于无监督3D目标检测任务。
This study introduces an unsupervised 3D object detection method named DOtA, which leverages internally shared information from multi-agent LiDAR scans for training without requiring external labels. DOtA comprises three modules: preliminary label generation, multi-scale bounding box encoding-based label filtering, and intra-label contrastive learning. This method has been tested on the V2V4Real and OPV2V datasets, demonstrating its superiority. The specific details of these datasets are not elaborated in this paper, but it can be learned from the provided datasets that they involve multi-agent collaborative perception and are used for unsupervised 3D object detection tasks.

- 1Learning to Detect Objects from Multi-Agent LiDAR Scans without Manual Labels厦门大学 · 2025年



