D2-Net: A Trainable CNN for Joint Detection and Description of Local Features

Mihai Dusmanu, Ignacio Rocco, Tomas Pajdla, Marc Pollefeys, Josef Sivic +2 more
5/9/2019
cs.CV

Official code: mihaidusmanu/d2-net (848 stars, PyTorch, last commit 2024-04-08).

Abstract

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.

View on arXivView PDF

Code Implementations(1)

mihaidusmanu/d2-netOfficial100%
PyTorch

D2-Net: A Trainable CNN for Joint Description and Detection of Local Features

848170Shell, MATLABApr 5, 20192 years agoNOASSERTION
cnncvpr2019local-featurespytorchvisual-localization

Similar Papers

XFeat: Accelerated Features for Lightweight Image Matching

86% match2 repos12024

HSCNet++: Hierarchical Scene Coordinate Classification and Regression for Visual Localization with Transformer

85% match2023

LoFTR: Detector-Free Local Feature Matching with Transformers

84% match1 repo3.0k2021

Self-supervising Fine-grained Region Similarities for Large-scale Image Localization

84% match1 repo2792020

ASLFeat: Learning Local Features of Accurate Shape and Localization

84% match1 repo3182020

Reinforced Feature Points: Optimizing Feature Detection and Description for a High-Level Task

84% match1 repo822019

Cite this paper

@article{dusmanu2019dnet,
  title  = {D2-Net: A Trainable CNN for Joint Detection and Description of Local Features},
  author = {Mihai Dusmanu and Ignacio Rocco and Tomas Pajdla and Marc Pollefeys and Josef Sivic and Akihiko Torii and Torsten Sattler},
  year   = {2019},
  eprint = {1905.03561},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1905.03561}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1905.03561.svg)](https://codeofpaper.com/paper/1905.03561)

Discussion