UGNA-VPR
收藏资源简介:
UGNA-VPR数据集是由国防科技大学智能科学和技术学院等机构收集的,适用于三维重建和视觉定位识别的任务。该数据集通过NeRF技术生成新的视点观测数据,以增强现有数据集的多视角多样性。数据集的创建是通过训练NeRF网络,然后使用自监督的不确定性估计网络识别具有高不确定性的地方,再利用NeRF生成新的合成观测数据用于VPR网络的进一步训练。该数据集旨在解决视觉定位识别中多方向驾驶或特征稀疏场景下识别准确率降低的问题。
The UGNA-VPR dataset was collected by institutions including the College of Intelligence Science and Technology, National University of Defense Technology, and is tailored for tasks of 3D reconstruction and visual place recognition (VPR). To enhance the multi-view diversity of existing datasets, this dataset generates novel viewpoint observation data via NeRF technology. The construction of the dataset involves first training a NeRF network, then utilizing a self-supervised uncertainty estimation network to identify locations with high uncertainty, and finally generating new synthetic observation data through NeRF for further training of VPR networks. This dataset aims to address the problem of reduced recognition accuracy in visual place recognition tasks under scenarios such as multi-directional driving or sparse feature distributions.




