D2-Net: A Trainable CNN for Joint Description and Detection of Local Features
Paper in proceeding, 2019
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 traditional counterparts based on early detection of low-level structures. We show that this model can be trained using pixel correspondences extracted from readily available large-scale SfM reconstructions, without any further annotations. The proposed method obtains state-of-the-art performance on both the difficult Aachen Day-Night localization dataset and the InLoc indoor localization benchmark, as well as competitive performance on other benchmarks for image matching and 3D reconstruction.
visual localization
3D reconstruction
machine learning
deep learning
local features
Author
Mihai Dusmanu
Ecole Normale Superieure (ENS)
Swiss Federal Institute of Technology in Zürich (ETH)
Institut National de Recherche en Informatique et en Automatique (INRIA)
Ignacio Rocco
Ecole Normale Superieure (ENS)
Institut National de Recherche en Informatique et en Automatique (INRIA)
Tomas Pajdla
Czech Technical University in Prague
Marc Pollefeys
Swiss Federal Institute of Technology in Zürich (ETH)
Microsoft Corporation
Josef Sivic
Institut National de Recherche en Informatique et en Automatique (INRIA)
Czech Technical University in Prague
Ecole Normale Superieure (ENS)
Akihiko Torii
Tokyo Institute of Technology
Torsten Sattler
Chalmers, Electrical Engineering, Signal Processing and Biomedical Engineering
Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
10636919 (ISSN)
Vol. 2019-June 8084-8093978-1-7281-3293-8 (ISBN)
IEEE / CVF Conference on Computer Vision and Pattern Recognition
Long Beach, USA,
Long Beach, USA,
Subject Categories (SSIF 2011)
Robotics
Computer Vision and Robotics (Autonomous Systems)
DOI
10.1109/CVPR.2019.00828