LF-Net: Learning Local Features from Images

TL;DR

LF-Net introduces a deep learning approach to learn local image features from collections of images without human supervision, leveraging depth and pose cues to outperform existing methods in feature matching at high speed.

Contribution

It proposes a novel deep architecture and training strategy that learns local features from images without manual labels, using depth and pose information for supervision.

Findings

Outperforms state-of-the-art on sparse feature matching

Operates at over 60 fps for QVGA images

Effective on both indoor and outdoor datasets

Abstract

We present a novel deep architecture and a training strategy to learn a local feature pipeline from scratch, using collections of images without the need for human supervision. To do so we exploit depth and relative camera pose cues to create a virtual target that the network should achieve on one image, provided the outputs of the network for the other image. While this process is inherently non-differentiable, we show that we can optimize the network in a two-branch setup by confining it to one branch, while preserving differentiability in the other. We train our method on both indoor and outdoor datasets, with depth data from 3D sensors for the former, and depth estimates from an off-the-shelf Structure-from-Motion solution for the latter. Our models outperform the state of the art on sparse feature matching on both datasets, while running at 60+ fps for QVGA images.