Robust Image Retrieval-based Visual Localization using Kapture

“…We compare our retrieval method to AP-GeM (Revaud et al, 2019a) and HOW (Tolias et al, 2020) and report results in Table 7. AP-GeM is the default method used in Kapture (Humenberger et al, 2020). We observe that using FIRe leads to better visual localization, specially in the most challenging scenario of night image localization and the strictest localization threshold: Performance improves by 2% compared to AP-GeM and by 1.5% compared to HOW on night images at a threshold of 0.25m and 2°.…”

mentioning

confidence: 88%

“…In this section, we evaluate FIRe for the task of visual localization, where retrieval is used as a first-stage filtering and before more precise, local feature-based geometric matching. To this end, we follow the pipeline proposed by Kapture 6 (Humenberger et al, 2020) on the Aachen Day-Night v1.1 dataset (Sattler et al, 2018). In this scenario, a global Structure-from-Motion map is built Retrieval method Day images Night images 0.25m, 2°0.5m, 5°5m, 10°0.25m, 2°0.5m, 5°5m, 10°A P-GeM (Revaud et al, 2019a) 88.8 96.6 99.6 72.3 86.9 97.9 HOW (Tolias et al, 2020) 90 Percentage of successfully localized images on the Aachen Day-Night v1.1 dataset when changing the retrieval method in the Kapture pipeline from Humenberger et al (2020).…”

Section: Application To Visual Localizationmentioning

confidence: 99%

Learning Super-Features for Image Retrieval

Weinzaepfel¹,

Lucas²,

Larlus³

et al. 2022

Preprint

View full text Add to dashboard Cite

Methods that combine local and global features have recently shown excellent performance on multiple challenging deep image retrieval benchmarks, but their use of local features raises at least two issues. First, these local features simply boil down to the localized map activations of a neural network, and hence can be extremely redundant. Second, they are typically trained with a global loss that only acts on top of an aggregation of local features; by contrast, testing is based on local feature matching, which creates a discrepancy between training and testing. In this paper, we propose a novel architecture for deep image retrieval, based solely on mid-level features that we call Super-features. These Super-features are constructed by an iterative attention module and constitute an ordered set in which each element focuses on a localized and discriminant image pattern. For training, they require only image labels. A contrastive loss operates directly at the level of Super-features and focuses on those that match across images. A second complementary loss encourages diversity. Experiments on common landmark retrieval benchmarks validate that Super-features substantially outperform state-of-the-art methods when using the same number of features, and only require a significantly smaller memory footprint to match their performance. Code and models are available at: https://github.com/naver/FIRe.

show abstract

“…RGB-D variants of scene coordinate regression methods dominate rankings for indoor re-localisation, which has been attributed to the inherent difficulty of the indoor scenario regarding texture-less surfaces and ambiguous structures that make it difficult to find and match sparse features [5,38,71,83]. For outdoor re-localisation, classical approaches, which match hand-crafted [61,71,81] or learned descriptors [21,32,57,58] at sparse feature locations to a 3D SfM reconstruction, achieve vastly superior results compared to scene coordinate regression. This has been attributed to an inability of scene coordinate regression to scale to spatially large scenes [62,77].…”

Section: Related Workmentioning

confidence: 99%

“…hLoc [57] combines image retrieval with Su-perPoint [21] features and SuperGlue [58] for matching, followed by P3P+RANSAC-based pose estimation. Den-seVLAD+R2D2 [32,55,78] uses DenseVLAD [78] for retrieving image pairs and R2D2 features for matching. The training images and poses are used to construct a 3D SfM map, and test images are localised using 2D-3D matches and P3P+RANSAC.…”

Section: Re-localisation Evaluationmentioning

confidence: 99%

On the Limits of Pseudo Ground Truth in Visual Camera Re-localisation

Brachmann¹,

Humenberger²,

Rother³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Benchmark datasets that measure camera pose accuracy have driven progress in visual re-localisation research. To obtain poses for thousands of images, it is common to use a reference algorithm to generate pseudo ground truth. Popular choices include Structure-from-Motion (SfM) and Simultaneous-Localisation-and-Mapping (SLAM) using additional sensors like depth cameras if available. Re-localisation benchmarks thus measure how well each method replicates the results of the reference algorithm. This begs the question whether the choice of the reference algorithm favours a certain family of re-localisation methods. This paper analyzes two widely used re-localisation datasets and shows that evaluation outcomes indeed vary with the choice of the reference algorithm. We thus question common beliefs in the re-localisation literature, namely that learning-based scene coordinate regression outperforms classical feature-based methods, and that RGB-Dbased methods outperform RGB-based methods. We argue that any claims on ranking re-localisation methods should take the type of the reference algorithm, and the similarity of the methods to the reference algorithm, into account.

show abstract

Robust Image Retrieval-based Visual Localization using Kapture

Cited by 15 publications

References 50 publications

MegLoc: A Robust and Accurate Visual Localization Pipeline

MegLoc: A Robust and Accurate Visual Localization Pipeline

Learning Super-Features for Image Retrieval

On the Limits of Pseudo Ground Truth in Visual Camera Re-localisation

Contact Info

Product

Resources

About