arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2003.10151cs.CV
收藏

GeoGraph:端到端几何线索的学习基于图的多视图目标检测

GeoGraph: Learning graph-based multi-view object detection with geometric cues end-to-end

IRISA, Université Bretagne Sud(布列塔尼南部大学IRISA研究所);ETH Zurich(苏黎世联邦理工学院)

Ahmed Samy Nassar, Stefano D'Aronco, Sébastien Lefèvre, Jan D. Wegner

摘要:

在本文中,我们提出了一种端到端的可学习方法,从多个视图中检测静态城市对象,重新识别实例,并最终为每个对象分配地理位置。我们的方法依赖于一个图神经网络(GNN),检测所有的对象,并输出其地理位置给定的图像和近似的相机姿态作为输入。我们的GNN同时对相对姿态和图像证据进行建模,并且还能够处理任意数量的输入视图。我们的方法是强大的闭塞,与相邻对象的外观相似,并通过联合推理视觉图像外观和相对姿态的观点发生严重变化。在两个具有挑战性的大规模数据集上进行的实验评估以及与最先进方法的比较表明,在准确性和效率方面都有显着和系统性的改进,检测和re-ID平均精度提高了2-6%,训练时间减少了8倍。

英文摘要:

In this paper we propose an end-to-end learnable approach that detects static urban objects from multiple views, re-identifies instances, and finally assigns a geographic position per object. Our method relies on a Graph Neural Network (GNN) to, detect all objects and output their geographic positions given images and approximate camera poses as input. Our GNN simultaneously models relative pose and image evidence, and is further able to deal with an arbitrary number of input views. Our method is robust to occlusion, with similar appearance of neighboring objects, and severe changes in viewpoints by jointly reasoning about visual image appearance and relative pose. Experimental evaluation on two challenging, large-scale datasets and comparison with state-of-the-art methods show significant and systematic improvements both in accuracy and efficiency, with 2-6% gain in detection and re-ID average precision as well as 8x reduction of training time.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑