arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2104.00680cs.CVcs.RO
收藏

LoFTR:基于Transformers的无检测器局部特征匹配

LoFTR: Detector-Free Local Feature Matching with Transformers

Zhejiang University(浙江大学);SenseTime Research(商汤科技研究院)

Jiaming Sun, Zehong Shen, Yuang Wang, Hujun Bao, Xiaowei Zhou

摘要:

提出了一种新的局部图像特征匹配方法。而不是执行图像特征检测,描述和匹配顺序,我们建议首先建立像素密集的匹配在一个粗略的水平,后来完善的好匹配在一个精细的水平。与使用成本量来搜索对应关系的密集方法相比,我们在Transformer中使用自我和交叉注意层来获得以两幅图像为条件的特征描述符。Transformer提供的全局感受野使我们的方法能够在低纹理区域产生密集的匹配,在这些区域,特征检测器通常很难产生可重复的兴趣点。在室内和室外数据集上的实验表明,LoFTR的性能大大优于最先进的方法。LoFTR在已发表的方法中,在视觉定位的两个公共基准中排名第一。

英文摘要:

We present a novel method for local image feature matching. Instead of performing image feature detection, description, and matching sequentially, we propose to first establish pixel-wise dense matches at a coarse level and later refine the good matches at a fine level. In contrast to dense methods that use a cost volume to search correspondences, we use self and cross attention layers in Transformer to obtain feature descriptors that are conditioned on both images. The global receptive field provided by Transformer enables our method to produce dense matches in low-texture areas, where feature detectors usually struggle to produce repeatable interest points. The experiments on indoor and outdoor datasets show that LoFTR outperforms state-of-the-art methods by a large margin. LoFTR also ranks first on two public benchmarks of visual localization among the published methods.

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。

↑