D2-Net: A Trainable Cnn For Joint Description And Detection Of Local Features

In this work we address the problem of finding reliable pixel-level correspondences under difficult imaging conditions. We propose an approach where a single convolutional neural network plays a dual role: It is simultaneously a dense feature descriptor and a feature detector By postponing the detection to a later stage, the obtained keypoints are more stable than their traditionalcounterparts based on early detection of low-level structures. We show that this model can be trainedusing pixel correspondences extractedfrom readilyavailable large-scaleSfM reconstructions, without any further annotations. The proposed method obtains state-of-the-artperformance on both the difficult Aachen Day-Night localizationdataset and the InLoc indoor localization benchmark, as well as competitive performance on other benchmarksfor image matching and 3D reconstruction.