Long-Term Visual Localization Revisited

Abstract

Visual localization enables autonomous vehicles to navigate in their surroundings and augmented reality applications to link virtual to real worlds. Practical visual localization approaches need to be robust to a wide variety of viewing conditions, including day-night changes, as well as weather and seasonal variations, while providing highly accurate six degree-of-freedom (6DOF) camera pose estimates. In this paper, we extend three publicly available datasets containing images captured under a wide variety of viewing conditions, but lacking camera pose information, with ground truth pose information, making evaluation of the impact of various factors on 6DOF camera pose estimation accuracy possible. We also discuss the performance of state-of-the-art localization approaches on these datasets. Additionally, we release around half of the poses for all conditions, and keep the remaining half private as a test set, in the hopes that this will stimulate research on long-term visual localization, learned local image features, and related research areas. Our datasets are available at visuallocalization.net, where we are also hosting a benchmarking server for automatic evaluation of results on the test set. The presented state-of-the-art results are to a large degree based on submissions to our server.

Rigidity Preserving Image Transformations and Equivariance in Perspective

“…Dataset. We use the CMU Extended Seasons dataset [2,54]. This dataset is collected using car-mounted cameras, producing images with a significant amount of radial distortion.…”

Scale-Net: Learning to Reduce Scale Differences for Large-Scale Invariant Image Matching

“…SIFT is a handcrafted feature detection algorithm whose robustness to scale changes is widely recognized. ASLFeat is a learningbased detect-and-describe feature extraction method which performs well on several broadly-used benchmarks [4], [50], [52]. GIFT-SP is a learning-based detect-then-describe feature extraction method, which uses Superpoint as the keypoint detector and uses GIFT as the descriptor and is robust to scale changes.…”

“…E STABLISHING pixel-level correspondences between two images is an essential basis for a wide range of computer vision tasks such as structure from motion [1], [2], augmented reality [3], visual localization [4], and Simultaneous localization and mapping (SLAM) [5]. Such correspondences are usually estimated by sparse local feature extraction and matching [6]- [16].…”

Decoupling Makes Weakly Supervised Local Feature Better

“…Specifically, the description network and detection network are combined into a single CNN. By combining the detection and description steps for joint optimization, the joint describe-then-detect pipeline achieves better performance than the detect-then-describe pipeline, especially under challenging conditions [42], [16]. However, these methods are fully supervised and rely on dense groundtruth correspondence labels for training.…”