MS2DG-Net: Progressive Correspondence Learning via Multiple Sparse Semantics Dynamic Graph

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Journal Year: 2022, Volume and Issue: unknown, P. 8963 - 8972

Published: June 1, 2022

Establishing superior-quality correspondences in an image pair is pivotal to many subsequent computer vision tasks. Using Euclidean distance between find neighbors and extract local information a common strategy previous works. However, most such works ignore similar sparse semantics two given images cannot capture topology among well. Therefore, deal with the above problems, Multiple Sparse Semantics Dynamic Graph Network (MS ² DG-Net) proposed, this paper, predict probabilities of as inliers recover camera poses. MS2 DG-Net dynamically builds graphs based on similarity images, correspondences, while maintaining permutation-equivariant. Extensive experiments prove that outperforms state-of-the-art methods outlier removal pose estimation tasks public datasets heavy outliers. Source code:https://github.com/changcaiyang/MS2DG-Net

Language: Английский

---

ALIKED: A Lighter Keypoint and Descriptor Extraction Network via Deformable Transformation

IEEE Transactions on Instrumentation and Measurement, Journal Year: 2023, Volume and Issue: 72, P. 1 - 16

Published: Jan. 1, 2023

Image keypoints and descriptors play a crucial role in many visual measurement tasks. In recent years, deep neural networks have been widely used to improve the performance of keypoint descriptor extraction. However, conventional convolution operations do not provide geometric invariance required for descriptor. To address this issue, we propose Sparse Deformable Descriptor Head (SDDH), which learns deformable positions supporting features each constructs descriptors. Furthermore, SDDH extracts at sparse instead dense map, enables efficient extraction with strong expressiveness. addition, relax reprojection error (NRE) loss from train extracted Experimental results show that proposed network is both powerful various tasks, including image matching, 3D reconstruction, relocalization.

Language: Английский

---

Robust image matching via local graph structure consensus

Pattern Recognition, Journal Year: 2022, Volume and Issue: 126, P. 108588 - 108588

Published: Feb. 13, 2022

Language: Английский

---

Multi-UAV Collaborative Absolute Vision Positioning and Navigation: A Survey and Discussion

Drones, Journal Year: 2023, Volume and Issue: 7(4), P. 261 - 261

Published: April 11, 2023

The employment of unmanned aerial vehicles (UAVs) has greatly facilitated the lives humans. Due to mass manufacturing consumer and support related scientific research, it can now be used in lighting shows, jungle search-and-rescues, topographical mapping, disaster monitoring, sports event broadcasting, among many other disciplines. Some applications have stricter requirements for autonomous positioning capability UAV clusters, requiring its precision within cognitive range a human or machine. Global Navigation Satellite System (GNSS) is currently only method that applied directly consistently positioning. Even with dependable GNSS, large-scale clustering drones might fail, resulting drone cluster bombardment. As type passive sensor, visual sensor compact size, low cost, wealth information, strong positional autonomy reliability, high accuracy. This automated navigation technology ideal swarms. application vision sensors collaborative task multiple UAVs effectively avoid interruption deficiency caused by factors such as field-of-view obstruction flight height limitation single achieve large-area group complex environments. paper examines (UAV navigation, distributed measurement fusion under dynamic topology, based on active behavior control multi-source sensing information). Current research constraints are compared appraised, most pressing issues addressed future anticipated researched. Through analysis discussion, been concluded integrated aforementioned methodologies aids enhancing cooperative capabilities during GNSS denial.

Language: Английский

---

SiLK: Simple Learned Keypoints

2021 IEEE/CVF International Conference on Computer Vision (ICCV), Journal Year: 2023, Volume and Issue: unknown, P. 22442 - 22451

Published: Oct. 1, 2023

Keypoint detection & descriptors are foundational technologies for computer vision tasks like image matching, 3D reconstruction and visual odometry. Hand-engineered methods Harris corners, SIFT, HOG have been used decades; more recently, there has a trend to introduce learning in an attempt improve key-point detectors. On inspection however, the results difficult interpret; recent learning-based employ vast diversity of experimental setups design choices: empirical often reported using different backbones, protocols, datasets, types supervisions or tasks. Since these differences coupled together, it raises natural question on what makes good learned keypoint detector. In this work, we revisit existing detectors by deconstructing their methodologies identifying key components. We re-design each component from first-principle propose Simple Learned Keypoints (SiLK) that is fully-differentiable, lightweight, flexible. Despite its simplicity, SiLK advances new state-of-the-art Detection Repeatability Homography Estimation HPatches Point-Cloud Registration task ScanNet, achieves competitive performance camera pose estimation 2022 Image Matching Challenge ScanNet.

Language: Английский

---