Vision-based localization of an agent in a map is an important problem in
robotics and computer vision. In that context, localization by learning
matchable image features is gaining popularity due to recent advances in
machine learning. Features that uniquely describe the visual contents of images
have a wide range of applications, including image retrieval and understanding.
In this work, we propose a method that learns image features targeted for
image-retrieval-based localization. Retrieval-based localization has several
benefits, such as easy maintenance and quick computation. However, the
state-of-the-art features only provide visual similarity scores which do not
explicitly reveal the geometric distance between query and retrieved images.
Knowing this distance is highly desirable for accurate localization, especially
when the reference images are sparsely distributed in the scene. Therefore, we
propose a novel loss function for learning image features which are both
visually representative and geometrically relatable. This is achieved by
guiding the learning process such that the feature and geometric distances
between images are directly proportional. In our experiments we show that our
features not only offer significantly better localization accuracy, but also
allow to estimate the trajectory of a query sequence in absence of the
reference images.

本文介绍了一种机器学习方法，用于学习基于图像检索的定位图像特征，通过引导学习过程，使图像特征与几何距离成比例，并在实验中表现出更好的定位准确度以及在缺乏参考图像的情况下估计查询序列轨迹的能力。

可几何映射的图像特征

Geometrically Mappable Image Features

We present an improved determination of the Hubble constant (H0) from Hubble
Space Telescope (HST) observations of 70 long-period Cepheids in the Large
Magellanic Cloud. These were obtained with the same WFC3 photometric system
used to measure Cepheids in the hosts of Type Ia supernovae. Gyroscopic control
of HST was employed to reduce overheads while collecting a large sample of
widely-separated Cepheids. The Cepheid Period-Luminosity relation provides a
zeropoint-free link with 0.4% precision between the new 1.2% geometric distance
to the LMC from Detached Eclipsing Binaries (DEBs) measured by Pietrzynski et
al (2019) and the luminosity of SNe Ia. Measurements and analysis of the LMC
Cepheids were completed prior to knowledge of the new LMC distance. Combined
with a refined calibration of the count-rate linearity of WFC3-IR with 0.1%
precision (Riess et al 2019), these three improved elements together reduce the
full uncertainty in the LMC geometric calibration of the Cepheid distance
ladder from 2.5% to 1.3%. Using only the LMC DEBs to calibrate the ladder we
find H0=74.22 +/- 1.82 km/s/Mpc including systematic uncertainties, 3% higher
than before for this particular anchor. Combining the LMC DEBs, masers in NGC
4258 and Milky Way parallaxes yields our best estimate: H0 = 74.03 +/- 1.42
km/s/Mpc, including systematics, an uncertainty of 1.91%---15% lower than our
best previous result. Removing any one of these anchors changes H0 by < 0.7%.
The difference between H0 measured locally and the value inferred from Planck
CMB+LCDM is 6.6+/-1.5 km/s/Mpc or 4.4 sigma (P=99.999% for Gaussian errors) in
significance, raising the discrepancy beyond a plausible level of chance. We
summarize independent tests which show this discrepancy is not readily
attributable to an error in any one source or measurement, increasing the odds
that it results from a cosmological feature beyond LambdaCDM.

使用恒星 Cepheids 測定了銀暈系統中大麥哲倫星雲（LMC）的橫距離，並且探索宇宙學中除了 LambdaCDM 殘余中的其他可能性因素

大麦哲伦星系的西佛星标准为测定哈勃常数提供了 1% 的基础，并提供了更强的超出 LambdaCDM 模型物理学的证据

Large Magellanic Cloud Cepheid Standards Provide a 1% Foundation for the  Determination of the Hubble Constant and Stronger Evidence for Physics Beyond  LambdaCDM