58 Hits in 2.8 sec

Wasserstein Distances for Stereo Disparity Estimation [article]

Divyansh Garg, Yan Wang, Bharath Hariharan, Mark Campbell, Kilian Q. Weinberger, Wei-Lun Chao
2021 arXiv   pre-print
We validate our approach on a variety of tasks, including stereo disparity and depth estimation, and the downstream 3D object detection.  ...  Existing approaches to depth or disparity estimation output a distribution over a set of pre-defined discrete values.  ...  It is compatible with most existing stereo depth or disparity estimation approaches -we only need to add an additional offset branch and replace the commonly used regression loss by the Wasserstein distance  ... 
arXiv:2007.03085v2 fatcat:cjdg3swjuve3xlhqpryv6ll6ju

Motion Language of Stereo Image Sequence

Tomoya Kato, Hayato Itoh, Atsushi Imiya
2017 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)  
These two operations extract a string of words of motion in front of the robots from stereo images captured by a robot-mounted camera system.  ...  This paper proposes a method for the symbolisation of temporal changes of the environments around the autonomous robot in a work space using the scene flow field for recognition of events.  ...  The KLT with piecewise-linear and locally quadratic constrains were used for the computation of stereo disparities between stereo pair images and optical flow between frames in each image sequence.  ... 
doi:10.1109/cvprw.2017.160 dblp:conf/cvpr/KatoII17 fatcat:26md73nm7bcazgabhpusuhy2tq

Environment reconstruction on depth images using Generative Adversarial Networks [article]

Lucas P. N. Matias, Jefferson R. Souza, Denis F. Wolf
2019 arXiv   pre-print
Our final contribution is a loss function focused on disparity data and a GAN able to extract depth features and estimate depth data by inpainting disparity images.  ...  Stereo cameras collect environment information at many levels, e.g., depth, color, texture, shape, which guarantee ample knowledge about the surroundings.  ...  [3] introduce their WGAN loss function based on the Wasserstein distance metric.  ... 
arXiv:1912.03992v1 fatcat:27s4w2y4ljfe5k43bsdzkio7me

SDF-MAN: Semi-Supervised Disparity Fusion with Multi-Scale Adversarial Networks

Can Pu, Runzi Song, Radim Tylecek, Nanbo Li, Robert Fisher
2019 Remote Sensing  
Assuming a Markov Random Field for the refined disparity map produces better estimates of the true disparity distribution.  ...  Uncertainty estimation and complex disparity relationships among pixels limit the accuracy and robustness of existing methods and there is no standard method for fusion of different kinds of depth data  ...  We thank Chengyang Zhao, Marija Jegorova and Timothy Hospedales for giving us good advice on this paper.  ... 
doi:10.3390/rs11050487 fatcat:s3yp4j72bzen3gaywsie5hkwpm

A Conditional Adversarial Network for Scene Flow Estimation [article]

Ravi Kumar Thakur, Snehasis Mukherjee
2019 arXiv   pre-print
The proposed network is the first attempt to estimate scene flow using generative adversarial networks, and is able to estimate both the optical flow and disparity from the input stereo images simultaneously  ...  We propose a conditional adversarial network SceneFlowGAN for scene flow estimation. The proposed SceneFlowGAN uses loss function at two ends: both generator and descriptor ends.  ...  The error in disparity is given by L1 loss. We use wasserstein metric as GAN loss function for stable training using gradient descent [1] .  ... 
arXiv:1904.11163v1 fatcat:ztnbw3quxzcohoiqvx7dctml5i

ITSA: An Information-Theoretic Approach to Automatic Shortcut Avoidance and Domain Generalization in Stereo Matching Networks [article]

WeiQin Chuah, Ruwan Tennakoon, Reza Hoseinnezhad, Alireza Bab-Hadiashar, David Suter
2022 arXiv   pre-print
Importantly, the proposed method enhances the robustness of the synthetic trained networks to the point that they outperform their fine-tuned counterparts (on real data) for challenging out-of-domain stereo  ...  State-of-the-art stereo matching networks trained only on synthetic data often fail to generalize to more challenging real data domains.  ...  The corresponding EPE is displayed on the estimated disparity map.  ... 
arXiv:2201.02263v2 fatcat:qdzmfql2mratzp3al43xl47sva

PLUMENet: Efficient 3D Object Detection from Stereo Images [article]

Yan Wang, Bin Yang, Rui Hu, Ming Liang, Raquel Urtasun
2021 arXiv   pre-print
While many approaches rely on expensive 3D sensors such as LiDAR to produce accurate 3D estimates, methods that exploit stereo cameras have recently shown promising results at a lower cost.  ...  Existing approaches tackle this problem in two steps: first depth estimation from stereo images is performed to produce a pseudo LiDAR point cloud, which is then used as input to a 3D object detector.  ...  [15] proposes a new loss function based on the Wasserstein distance to further improve the depth prediction on object boundaries. b) LiDAR-based 3D Object Detection: LiDAR sensors capture high quality  ... 
arXiv:2101.06594v3 fatcat:spdwihruxney5omg23w54gszky

On the benefit of adversarial training for monocular depth estimation

Rick Groenendijk, Sezer Karaoglu, Theo Gevers, Thomas Mensink
2019 Computer Vision and Image Understanding  
For the quality of the image reconstruction and disparity prediction, a combination of different losses is used, including L1 image reconstruction losses and left-right disparity smoothness.  ...  A model can be trained in a self-supervised setting on stereo pairs of images, where depth (disparities) are an intermediate result in a right-to-left image reconstruction pipeline.  ...  Acknowledgement This research was supported in part by the Dutch Organisation for Scientific Research, Netherlands via the VENI grant What & Where awarded to Dr. Mensink.  ... 
doi:10.1016/j.cviu.2019.102848 fatcat:pzycd5ynubfglbpcfhyji7c5lu

DDL-MVS: Depth Discontinuity Learning for MVS Networks [article]

Nail Ibrahimli, Hugo Ledoux, Julian Kooij, Liangliang Nan
2022 arXiv   pre-print
Our idea is to jointly estimate the depth and boundary maps where the boundary maps are explicitly used for further refinement of the depth maps.  ...  Traditional MVS methods have good accuracy but struggle with completeness, while recently developed learning-based multi-view stereo (MVS) techniques have improved completeness except accuracy being compromised  ...  The continuous disparity network [14] aims to regress the multi-modal depth by jointly estimating both probability and offset volume by minimizing a Wasserstein distance between the ground truth and  ... 
arXiv:2203.01391v2 fatcat:wkuqjxurdzdk3ajb4qyzhfzmeq

Towards Continual, Online, Self-Supervised Depth [article]

Muhammad Umar Karim Khan
2022 arXiv   pre-print
Effort has been made to make the proposed approach suitable for practical use. We apply our method to both structure-from-motion and stereo depth estimation.  ...  Qualitative and quantitative results with both structure-from-motion and stereo show superior forgetting as well as adaptation performance compared to recent methods.  ...  Wasserstein distance is used in [8] for depth estimation and 3D object detection. Authors in [9] propose a self-supervised scheme to estimate the depth and 3D scene flow simultaneously.  ... 
arXiv:2103.00369v3 fatcat:jhzpki73zvczdp5z2npw3xsbtu

Relativistic Approach for Training Self-supervised Adversarial Depth Prediction Model using Symmetric Consistency

Hyeseung Park, Seungchul Park, Younbok Joo
2020 IEEE Access  
[15] viewed monocular depth estimation as a stereo matching problem.  ...  They utilize the deep stereo matching network as a proxy to learn depth from synthetic data and provide dense supervision for training monocular depth estimation networks.  ... 
doi:10.1109/access.2020.3036893 fatcat:dhs6chiwvben3nwhqcahgvrtmy

DSGN++: Exploiting Visual-Spatial Relation for Stereo-based 3D Detectors [article]

Yilun Chen, Shijia Huang, Shu Liu, Bei Yu, Jiaya Jia
2022 arXiv   pre-print
We first revisit the prior stereo detector DSGN for its stereo volume construction ways for representing both 3D geometry and semantics.  ...  Second, for grasping differently spaced features, we present a novel stereo volume -- Dual-view Stereo Volume (DSV) that integrates front-view and top-view features and reconstructs sub-voxel depth in  ...  CDN [67] further refine depth prediction near object boundarys via a Wasserstein distance-based loss.  ... 
arXiv:2204.03039v3 fatcat:scjsfynei5b3fbs3ewxmsimfgy

LIGA-Stereo: Learning LiDAR Geometry Aware Representations for Stereo-based 3D Detector [article]

Xiaoyang Guo, Shaoshuai Shi, Xiaogang Wang, Hongsheng Li
2021 arXiv   pre-print
Stereo-based 3D detection aims at detecting 3D object bounding boxes from stereo images using intermediate depth maps or implicit 3D geometry representations, which provides a low-cost solution for 3D  ...  To solve the problem, we propose LIGA-Stereo (LiDAR Geometry Aware Stereo Detector) to learn stereo-based 3D detectors under the guidance of high-level geometry-aware representations of LiDAR-based detection  ...  Wasserstein In D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, and R. distances for stereo disparity estimation.  ... 
arXiv:2108.08258v1 fatcat:hz7ix7azmfbf3oli7mpay7jf3a

Panchromatic Image Super-Resolution Via Self Attention-Augmented Wasserstein Generative Adversarial Network

Juan Du, Kuanhong Cheng, Yue Yu, Dabao Wang, Huixin Zhou
2021 Sensors  
features for detail enhancement.  ...  To address this problem, an improved SR model which involves the self-attention augmented Wasserstein generative adversarial network ( SAA-WGAN) is designed to dig out the reference information among multiple  ...  Wang created a parallax-attention mechanism (PASSRnet [16] ) to integrate the information from a stereo image pair, handling different stereo images with large disparity variations.  ... 
doi:10.3390/s21062158 pmid:33808682 fatcat:c5lbzf5jqrbwvck7dxub6dania

Limited-angle tomographic reconstruction of dense layered objects by dynamical machine learning [article]

Iksung Kang, Alexandre Goy, George Barbastathis
2020 arXiv   pre-print
Recently, it was shown that one effective way to learn the priors for strongly scattering yet highly structured 3D objects, e.g. layered and Manhattan, is by a static neural network [Goy et al, Proc.  ...  Even though training used NPCC as in (12) , we investigated two additional metrics for testing: probability of error (PE), the Wasserstein distance [82, 83] .  ...  PE can be reduced to have a similar, but not equivalent, form to that of the Wasserstein distance.  ... 
arXiv:2007.10734v1 fatcat:kv37pc3vffcwbfyyzx2ycchkaq
« Previous Showing results 1 — 15 out of 58 results