Filters








1,345 Hits in 4.1 sec

Beyond Tracking: Selecting Memory and Refining Poses for Deep Visual Odometry

Fei Xue, Xin Wang, Shunkai Li, Qiuyuan Wang, Junqiu Wang, Hongbin Zha
2019 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
The Refining component ameliorates previous results with the contexts stored in the Memory by adopting a spatial-temporal attention mechanism for feature distilling.  ...  Most previous learning-based visual odometry (VO) methods take VO as a pure tracking problem.  ...  Unsupervised methods Mimicking the conventional structure from motion, SfmLearner [39] learns the single view depth and ego-motion from monocular image snippets using photometric errors as supervisory  ... 
doi:10.1109/cvpr.2019.00877 dblp:conf/cvpr/XueWLWWZ19 fatcat:kgc4ovzobbhcfdb4z4gqyibwm4

Deep incremental learning for efficient high-fidelity face tracking

Chenglei Wu, Takaaki Shiratori, Yaser Sheikh
2018 ACM Transactions on Graphics  
for the comparisons in Section 6.4, and Colin Lea and all reviewers for their constructive discussions and feedback.  ...  ACKNOWLEDGMENTS We would like to thank the actors for allowing us to use their data in Section 6.1, the authors of Beeler and colleagues [2011] and Fyffe and colleagues [2017] for providing their data  ...  Quantitative Evaluation with Synthetic Data We created a synthetic data from 1260 frames out of the tracking result of the A1-1 sequence for quantitative evaluation.  ... 
doi:10.1145/3272127.3275101 fatcat:4bsyon3xuzfuzjn43pupxaye44

Good Feature Selection for Least Squares Pose Optimization in VO/VSLAM [article]

Yipu Zhao, Patricio A. Vela
2019 arXiv   pre-print
Unlike existing feature selection works that are focused on efficiency only, our method significantly improves the accuracy of pose tracking, while introducing little overhead.  ...  Integrating Max-logDet feature selection into a state-of-the-art visual SLAM system leads to accuracy improvements with low overhead, as demonstrated via evaluation on a public benchmark.  ...  POSE TRACKING Base Feat.  ... 
arXiv:1905.07807v1 fatcat:imx5gsojfbdapiy5bls4s4jrnu

Visual Odometry Revisited: What Should Be Learnt? [article]

Huangying Zhan, Chamara Saroj Weerasekera, Jiawang Bian, Ian Reid
2020 arXiv   pre-print
More importantly, our system does not suffer from the scale-drift issue being aided by a scale consistent single-view depth CNN.  ...  Most existing VO/SLAM systems with superior performance are based on geometry and have to be carefully designed for different application scenarios.  ...  from synthetic to real.  ... 
arXiv:1909.09803v4 fatcat:2yf6ozwtmbgj3bcete3ob4qgt4

Beyond Tracking: Selecting Memory and Refining Poses for Deep Visual Odometry [article]

Fei Xue, Xin Wang, Shunkai Li, Qiuyuan Wang, Junqiu Wang, Hongbin Zha
2019 arXiv   pre-print
The Refining component ameliorates previous results with the contexts stored in the Memory by adopting a spatial-temporal attention mechanism for feature distilling.  ...  Most previous learning-based visual odometry (VO) methods take VO as a pure tracking problem.  ...  Unsupervised methods Mimicking the conventional structure from motion, SfmLearner [39] learns the single view depth and ego-motion from monocular image snippets using photometric errors as supervisory  ... 
arXiv:1904.01892v2 fatcat:y6w24oqebfeu3hs7fotm22rwhm

Guided Feature Selection for Deep Visual Odometry [article]

Fei Xue, Qiuyuan Wang, Xin Wang, Wei Dong, Junqiu Wang, Hongbin Zha
2018 arXiv   pre-print
Different from current monocular visual odometry methods, our approach is established on the intuition that features contribute discriminately to different motion patterns.  ...  To enhance the ability of feature selection, we further introduce an effective context-aware guidance mechanism to force each branch to distill related information for specific motion pattern explicitly  ...  Different from DeepVO [35] and ESP-VO [36] , we keep the structure of feature-maps for retaining the spatial formulation rather than compressing features into 1D vectors.  ... 
arXiv:1811.09935v1 fatcat:dt2yhqilhfbkvgetg6egyarrki

DF-VO: What Should Be Learnt for Visual Odometry? [article]

Huangying Zhan, Chamara Saroj Weerasekera, Jia-Wang Bian, Ravi Garg, Ian Reid
2021 arXiv   pre-print
More importantly, monocular methods suffer from scale-drift issue, i.e., errors accumulate over time.  ...  Multi-view geometry-based methods dominate the last few decades in monocular Visual Odometry for their superior performance, while they have been vulnerable to dynamic and low-texture scenes.  ...  Acknowledgment This work was supported by the UoA Scholarship to HZ, the ARC Laureate Fellowship FL130100102 to IR and the Australian Centre of Excellence for Robotic Vision CE140100016.  ... 
arXiv:2103.00933v1 fatcat:gs4bsysoozelfmv7mdntx3xiba

SuperPoint features in endoscopy [article]

O. L. Barbed, F. Chadebecq, J. Morlana, J.M. Martínez-Montiel, A. C. Murillo
2022 arXiv   pre-print
We explore the potential of the well known self-supervised approach SuperPoint, present an adapted variation for the endoscopic domain and propose a challenging evaluation framework.  ...  Local feature extraction and matching is a key step for many computer vision applications, specially regarding 3D modelling.  ...  The need for Structure-from-Motion labels to train supervised methods makes these approaches impractical for applications such as endoscopy.  ... 
arXiv:2203.04302v1 fatcat:yfe7ypdp55durhgknbzp43wtgq

Learned Camera Gain and Exposure Control for Improved Visual Feature Detection and Matching [article]

Justin Tomasi, Brandon Wagstaff, Steven L. Waslander, Jonathan Kelly
2021 arXiv   pre-print
higher number of inlier feature matches than competing camera parameter control algorithms.  ...  In this paper, we explore a data-driven approach to account for environmental lighting changes, improving the quality of images for use in visual odometry (VO) or visual simultaneous localization and mapping  ...  This behaviour is demonstrated before entering a tunnel (left column) and while exiting from a tunnel (right column). Inlier feature tracks are shown in red.  ... 
arXiv:2102.04341v1 fatcat:mhdkpiqm5vhqfnapmt6f53aoxe

Table of contents

2018 2018 International Conference on 3D Vision (3DV)  
Single View 360 Video 334 Sam Fowler (University of Surrey), Hansung Kim (University of Surrey), and Adrian Hilton (University of Surrey) FEATS: Synthetic Feature Tracks for Structure from Motion Evaluation  ...  (Amazon), and Bernt Schiele (MPI) Progressive Large-Scale Structure-from-Motion with Orthogonal MSTs 79 Hainan Cui (Chinese Academy of Sciences), Shuhan Shen (Chinese Academy of Sciences), Wei Gao  ... 
doi:10.1109/3dv.2018.00004 fatcat:zxjnidy5f5ai7aeednry3ggt7y

GN-Net: The Gauss-Newton Loss for Multi-Weather Relocalization [article]

Lukas von Stumberg, Patrick Wenzel, Qadeer Khan, Daniel Cremers
2019 arXiv   pre-print
Furthermore, we release an evaluation benchmark for relocalization tracking against different types of weather. Our benchmark is available at https://vision.in.tum.de/gn-net.  ...  To overcome this, we propose GN-Net: a network optimized with the novel Gauss-Newton loss for training weather invariant deep features, tailored for direct image alignment.  ...  Deep direct image alignment: BA-NET [26] introduces a network architecture to solve the structure from motion (SfM) problem via feature-metric bundle adjustment.  ... 
arXiv:1904.11932v3 fatcat:wy62zn7ui5aonm5ysboltui6ba

LIC-Fusion 2.0: LiDAR-Inertial-Camera Odometry with Sliding-Window Plane-Feature Tracking [article]

Xingxing Zuo, Yulin Yang, Patrick Geneva, Jiajun Lv, Yong Liu, Guoquan Huang, Marc Pollefeys
2020 arXiv   pre-print
A novel outlier rejection criterion is proposed in the plane-feature tracking for high-quality data association.  ...  ., LIC-Fusion 2.0), which introduces a novel sliding-window plane-feature tracking for efficiently processing 3D LiDAR point clouds.  ...  LiDAR Plane Feature Update Analogous to point features [14] , we divide all the tracked plane features from the LiDAR point clouds into "MSCKF" and "SLAM" based on the track length.  ... 
arXiv:2008.07196v1 fatcat:ihj6yz2fobaa7nifgctkje3nye

OAS-Net: Occlusion Aware Sampling Network for Accurate Optical Flow [article]

Lingtong Kong, Xiaohang Yang, Jie Yang
2021 arXiv   pre-print
to correlate with source feature for building 3D matching cost volume.  ...  Third, a shared flow and occlusion awareness decoder is adopted for structure compactness. Experiments on Sintel and KITTI datasets demonstrate the effectiveness of proposed approaches.  ...  Fig. 2 : 2 A single pyramid level for occlusion aware optical flow estimation in OAS-Net, feat 1 and feat 2 are pyramid features. Fig. 3 : 3 Comparison on Sintel Final test dataset.  ... 
arXiv:2102.00364v1 fatcat:cjz4p6zk4feohj7hk5tzswb5b4

Incorporating Learnt Local and Global Embeddings into Monocular Visual SLAM [article]

Huaiyang Huang, Haoyang Ye, Yuxiang Sun, Lujia Wang, Ming Liu
2021 arXiv   pre-print
Traditional approaches for Visual Simultaneous Localization and Mapping (VSLAM) rely on low-level vision information for state estimation, such as handcrafted local features or the image gradient.  ...  With a probabilistic explanation of keypoint prediction, we formulate the camera pose tracking in a direct manner and parameterize local features with uncertainty taken into account.  ...  optimizing the structure and camera motion.  ... 
arXiv:2108.02028v1 fatcat:s23gu5chjfhpjpjk7vrxu67hcu

Deep Learning in Video Multi-Object Tracking: A Survey

Gioele Ciaparrone, Francisco Luque Sánchez, Siham Tabik, Luigi Troiano, Roberto Tagliaferri, Francisco Herrera
2019 Neurocomputing  
The problem of Multiple Object Tracking (MOT) consists in following the trajectory of different objects in a sequence, usually a video.  ...  In recent years, with the rise of Deep Learning, the algorithms that provide a solution to this problem have benefited from the representational power of deep models.  ...  In each column, the approach for each paper in that step is shown. app. means appearance, mot. means motion, feat. means features, pred. means prediction; O and B in the Mode column indicate Online and  ... 
doi:10.1016/j.neucom.2019.11.023 fatcat:loiw353o4rgvpkl5sdpndodk4m
« Previous Showing results 1 — 15 out of 1,345 results