Filters








842 Hits in 3.8 sec

Learning 3D Semantic Reconstruction on Octrees

Xiaojuan Wang, Marc Pollefeys, Martin Oswald, Ian Cherabier
2019
We present a fully convolutional neural network for semantic 3D reconstruction by using octree representation.  ...  This master thesis project involves almost every aspect of the 3D neural network, from low-level implementation such as convolution on octree in GPU to high-level reconstruction performance evaluation.  ...  Focus of This Work In this work, we are interested in learning semantic 3D reconstruction on octrees, predicting both the semantics and the octree structure of the scene.  ... 
doi:10.3929/ethz-b-000332863 fatcat:2j3ywucwlbb4dihks5ijlqwo3m

Deep Octree-based CNNs with Output-Guided Skip Connections for 3D Shape and Scene Completion [article]

Peng-Shuai Wang and Yang Liu and Xin Tong
2020 arXiv   pre-print
Our network is built upon the octree-based CNNs (O-CNN) with U-Net like structures, which enjoys high computational and memory efficiency and supports to construct a very deep network structure for 3D  ...  We show that with these simple adaptions -- output-guided skip-connection and deeper O-CNN (up to 70 layers), our network achieves state-of-the-art results in 3D shape completion and semantic scene computation  ...  Experiments We evaluate our networks on the tasks of 3D shape completion and semantic scene completion.  ... 
arXiv:2006.03762v1 fatcat:sjakew3nl5d7xmeol4vrskcejy

Fusion-Aware Point Convolution for Online Semantic 3D Scene Segmentation [article]

Jiazhao Zhang, Chenyang Zhu, Lintao Zheng, Kai Xu
2022 arXiv   pre-print
Online semantic 3D segmentation in company with real-time RGB-D reconstruction poses special challenges such as how to perform 3D convolution directly over the progressively fused 3D geometric data, and  ...  We propose a novel fusion-aware 3D point convolution which operates directly on the geometric surface being reconstructed and exploits effectively the inter-frame correlation for high quality 3D feature  ...  We are also grateful to Quang-Hieu Pham for the help on comparison with ProgressiveFusion.  ... 
arXiv:2003.06233v4 fatcat:6cjtzrywzrguxagr4byprpo564

Fusion-Aware Point Convolution for Online Semantic 3D Scene Segmentation

Jiazhao Zhang, Chenyang Zhu, Lintao Zheng, Kai Xu
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
Online semantic 3D segmentation in company with realtime RGB-D reconstruction poses special challenges such as how to perform 3D convolution directly over the progressively fused 3D geometric data, and  ...  learning.  ...  We are also grateful to Quang-Hieu Pham for the help on comparison with ProgressiveFusion. This work  ... 
doi:10.1109/cvpr42600.2020.00459 dblp:conf/cvpr/Zhang0Z020 fatcat:rbxi2zfr3vfijhi27v6hdvpb6u

Deep Octree-based CNNs with Output-Guided Skip Connections for 3D Shape and Scene Completion

Peng-Shuai Wang, Yang Liu, Xin Tong
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)  
Our network is built upon the octree-based CNNs (O-CNN) with U-Net like structures, which enjoys high computational and memory efficiency and supports to construct a very deep network structure for 3D  ...  We show that with these simple adaptionsoutput-guided skip-connection and deeper O-CNN (up to 70 layers), our network achieves state-of-the-art results in 3D shape completion and semantic scene computation  ...  Experiments We evaluate our networks on the tasks of 3D shape completion and semantic scene completion.  ... 
doi:10.1109/cvprw50498.2020.00141 dblp:conf/cvpr/Wang0020 fatcat:zbb4pg6dzbdjtff3b2bt62ismi

OctNetFusion: Learning Depth Fusion from Data [article]

Gernot Riegler, Ali Osman Ulusoy, Horst Bischof, Andreas Geiger
2017 arXiv   pre-print
In this paper, we present a learning based approach to depth fusion, i.e., dense 3D reconstruction from multiple depth images.  ...  By learning the structure of real world 3D objects and scenes, our approach is further able to reconstruct occluded regions and to fill in gaps in the reconstruction.  ...  In this work, we do not consider the semantic class of the reconstructed object or scene and focus on the generic 3D reconstruction problem using a global model.  ... 
arXiv:1704.01047v3 fatcat:pthbqpunhzdovlu6bglllpatuq

OctSqueeze: Octree-Structured Entropy Model for LiDAR Compression

Lila Huang, Shenlong Wang, Kelvin Wong, Jerry Liu, Raquel Urtasun
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
We then design a tree-structured conditional entropy model that models the probabilities of the octree symbols to encode the octree into a compact bitstream.  ...  The results demonstrate that our approach reduces the bitrate by 10-20% at the same reconstruction quality, compared to the previous state-of-the-art.  ...  Octree-Structured Entropy Model In this work we tackle the problem of lossy compression on 3D LiDAR point clouds.  ... 
doi:10.1109/cvpr42600.2020.00139 dblp:conf/cvpr/HuangWWLU20 fatcat:2y2vyl2l2vepnarahdyzzhbovu

OctSqueeze: Octree-Structured Entropy Model for LiDAR Compression [article]

Lila Huang, Shenlong Wang, Kelvin Wong, Jerry Liu, Raquel Urtasun
2021 arXiv   pre-print
We then design a tree-structured conditional entropy model that models the probabilities of the octree symbols to encode the octree into a compact bitstream.  ...  The results demonstrate that our approach reduces the bitrate by 10-20% at the same reconstruction quality, compared to the previous state-of-the-art.  ...  In Fig. 2 and 3 , we show the reconstruction quality of our method versus Draco. Then, in Fig. 4 and 5, we show their respective downstream semantic segmentation performance.  ... 
arXiv:2005.07178v2 fatcat:okvdql2cqbap5pvhajfaz577fi

Octree Generating Networks: Efficient Convolutional Architectures for High-resolution 3D Outputs [article]

Maxim Tatarchenko, Alexey Dosovitskiy, Thomas Brox
2017 arXiv   pre-print
The network learns to predict both the structure of the octree, and the occupancy values of individual cells. This makes it a particularly valuable technique for generating 3D shapes.  ...  We present a deep convolutional decoder architecture that can generate volumetric 3D outputs in a compute- and memory-efficient manner by using an octree representation.  ...  We also thank Nikolaus Mayer for his help with 3D model visualization and manuscript preparation.  ... 
arXiv:1703.09438v3 fatcat:377szncxofexvoedu3svudxcry

OctNet: Learning Deep 3D Representations at High Resolutions

Gernot Riegler, Ali Osman Ulusoy, Andreas Geiger
2017 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)  
We present OctNet, a representation for deep learning with sparse 3D data.  ...  We demonstrate the utility of the proposed OctNet on three different problems involving three-dimensional data: 3D classification, 3D orientation estimation of unknown object instances and semantic segmentation  ...  We analyzed the importance of high resolution inputs on several 3D learning tasks, such as object categorization, pose estimation and semantic segmentation.  ... 
doi:10.1109/cvpr.2017.701 dblp:conf/cvpr/RieglerUG17 fatcat:b7ojbfogkbclho5u4iccsvyqpi

MuSCLE: Multi Sweep Compression of LiDAR using Deep Entropy Models [article]

Sourav Biswas, Jerry Liu, Kelvin Wong, Shenlong Wang, Raquel Urtasun
2021 arXiv   pre-print
We then use the learned probability to encode the full data stream into a compact one.  ...  Towards this goal, we propose a novel conditional entropy model that models the probabilities of the octree symbols by considering both coarse level geometry and previous sweeps' geometric and intensity  ...  space from the (t − 1) sweep at the same level as i, p i is the 3D position of each node, and σ denotes a learned MLP.  ... 
arXiv:2011.07590v2 fatcat:viyouar7grg7zgm25kvoczgw34

Large-Scale Semantic 3D Reconstruction: An Adaptive Multi-resolution Model for Multi-class Volumetric Labeling

Maros Blaha, Christoph Vogel, Audrey Richard, Jan D. Wegner, Thomas Pock, Konrad Schindler
2016 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)  
Given a set of images of a scene, semantic 3D reconstruction aims to densely reconstruct both the 3D shape of the scene and a segmentation into semantic object classes.  ...  We propose an adaptive multi-resolution formulation of semantic 3D reconstruction.  ...  Moreover, our basic idea is generic and not limited to semantic 3D reconstruction.  ... 
doi:10.1109/cvpr.2016.346 dblp:conf/cvpr/BlahaVRWPS16 fatcat:fi3egfnfdncz7axqn3h2ebhqke

Leaving Flatland: Realtime 3D Stereo Semantic Reconstruction [chapter]

Radu Bogdan Rusu, Aravind Sundaresan, Benoit Morisset, Motilal Agrawal, Michael Beetz
2008 Lecture Notes in Computer Science  
The 3D geometric model is used to infer different terrain types and construct a 3D semantic model which can be used for path planning or teleoperation.  ...  To validate our approach, we show results obtained on multiple datasets and perform a comparison with other similar initiatives.  ...  Acknowledgments The authors thank Edward Van Reuth and DARPA for support on the "Leaving Flatland" project (contract #FA8650-04-C-7136).  ... 
doi:10.1007/978-3-540-88513-9_99 fatcat:3z56f54sozgulpaedqtwchbngm

OctField: Hierarchical Implicit Functions for 3D Modeling [article]

Jia-Heng Tang, Weikai Chen, Jie Yang, Bo Wang, Songrun Liu, Bo Yang, Lin Gao
2021 arXiv   pre-print
We achieve this goal by introducing a hierarchical octree structure to adaptively subdivide the 3D space according to the surface occupancy and the richness of part geometry.  ...  We demonstrate the value of OctField for a range of shape modeling and reconstruction tasks, showing superiority over alternative approaches.  ...  We believe the structural information is crucial for improving the modeling accuracy and future applications(e.g. 3D semantic understanding, editing). Learning-based Generative Models.  ... 
arXiv:2111.01067v1 fatcat:l2d4t423srdwnbkkgd3rxtm4uy

Using Deep Learning in Semantic Classification for Point Cloud Data

Xuanxia Yao, Jia Guo, Juan Hu, QiXuan Cao
2019 IEEE Access  
The experiments show its competitive performance in many 3D tasks, such as object classification and semantic segmentation. INDEX TERMS Point cloud, PointNet, 3D deep learning, octree, neural network.  ...  Point cloud is an important 3D data structure, but its irregular format brings great challenges to deep learning.  ...  Currently, advances in 3D reconstruction [4] and 3D graphics [5] allow us to capture and model large amounts of 3D data.  ... 
doi:10.1109/access.2019.2905546 fatcat:7x4r7syaazhrpfqlamtu6ix3ga
« Previous Showing results 1 — 15 out of 842 results