323 Hits in 4.6 sec

How Powerful Are Randomly Initialized Pointcloud Set Functions? [article]

Aditya Sanghi, Pradeep Kumar Jayaraman
2020 arXiv   pre-print
We study random embeddings produced by untrained neural set functions, and show that they are powerful representations which well capture the input features for downstream tasks such as classification,  ...  We obtain surprising results that show that random set functions can often obtain close to or even better accuracy than fully trained models.  ...  In this work, we study randomly initialized set functions, particularly for pointclouds in a similar vein, and make the following contributions: First, we experiment with well known set-based neural networks  ... 
arXiv:2003.05410v1 fatcat:ttfl35wahve3tcv2lxq4mk34my

Intrinsic Neural Fields: Learning Functions on Manifolds [article]

Lukas Koestler, Daniel Grittner, Michael Moeller, Daniel Cremers, Zorah Lähner
2022 arXiv   pre-print
The extrinsic embedding ignores known intrinsic manifold properties and is inflexible wrt. transfer of the learned function.  ...  Some of their advantages are a sound theoretic foundation and an easy implementation in current deep learning frameworks.  ...  Specifically, potential background pixels are identified based on their intensity due to the mostly white background. They are removed if they are close to the boundary of the initial mask.  ... 
arXiv:2203.07967v3 fatcat:2rxv6eedp5cxzlp2beoiojswna

Texture Fields: Learning Texture Representations in Function Space [article]

Michael Oechsle, Lars Mescheder, Michael Niemeyer, Thilo Strauss, Andreas Geiger
2019 arXiv   pre-print
In this paper, we propose Texture Fields, a novel texture representation which is based on regressing a continuous 3D function parameterized with a neural network.  ...  We show that Texture Fields are able to represent high frequency texture and naturally blend with modern deep learning techniques.  ...  In the first part, we investigate the representation power of Texture Fields by analyzing how well the method can represent high frequency textures when trained on a single 3D object.  ... 
arXiv:1905.07259v1 fatcat:tsyq2cdhp5ghbgunry2fgocvfa

Real-Time Freespace Segmentation on Autonomous Robots for Detection of Obstacles and Drop-Offs [article]

Anish Singhani
2019 arXiv   pre-print
We then show how this freespace map can be used for real-time navigation on an indoor robot.  ...  The results show that our system generalizes well, is suitable for real-time operation, and runs at around 55 fps on a small indoor robot powered by a low-power embedded GPU.  ...  The following equation shows how the blur factor β is calculated, given that the function L(x, y) is a discrete convolution of the input image I(x, y) where x and y are image coordinates: L(x, y) =    ... 
arXiv:1902.00842v1 fatcat:hazqmamk7ze2dcnacrl37kxagy

A Deep Signed Directional Distance Function for Object Shape Representation [article]

Ehsan Zobeidi, Nikolay Atanasov
2021 arXiv   pre-print
Neural networks that map 3D coordinates to signed distance function (SDF) or occupancy values have enabled high-fidelity implicit representations of object shape.  ...  This paper develops a new shape model that allows synthesizing novel distance views by optimizing a continuous signed directional distance function (SDDF).  ...  For spheres with radii 0.4, 0.6, 10 the errors are 0.0076, 0.0268, 33.9721, respectively. First, we investigate how noise in the training data affects our method.  ... 
arXiv:2107.11024v2 fatcat:x2cbg4nenfc2nlgfo55qp6fo3y

3D3L: Deep Learned 3D Keypoint Detection and Description for LiDARs [article]

Dominic Streiff, Lukas Bernreiter, Florian Tschopp, Marius Fehr, Roland Siegwart
2021 arXiv   pre-print
Pointcloud registration methods working with unstructured pointclouds such as ICP are often computationally expensive or require a good initial guess.  ...  With the advent of powerful, light-weight 3D LiDARs, they have become the hearth of many navigation and SLAM algorithms on various autonomous systems.  ...  Therefore, we check how much the two pointclouds P and P overlap.  ... 
arXiv:2103.13808v2 fatcat:4izwxty2zvds3lct7scdu5usju

Supervised Learning of Labeled Pointcloud Differences via Cover-Tree Entropy Reduction [article]

Abraham Smith and Paul Bendich and John Harer and Alex Pieloch and Jay Hineman
2018 arXiv   pre-print
CDER applies to a training set of labeled pointclouds embedded in a common Euclidean space.  ...  If typical pointclouds corresponding to distinct labels tend to differ at any scale in any sub-region, CDER can identify these differences in (typically) linear time, creating a set of distributional coordinates  ...  The obvious questions are: how many Gaussians (modes) should be used, what are their means, and what are their covariances?  ... 
arXiv:1702.07959v3 fatcat:qdxf3ilmpfc2njtq5ozihs2biq

Shape Completion Enabled Robotic Grasping [article]

Jacob Varley, Chad DeChant, Adam Richardson, Joaquín Ruales, Peter Allen
2017 arXiv   pre-print
We explore how the quality of completions vary based on several factors.  ...  Runtime shape completion is very rapid because most of the computational costs of shape completion are borne during offline training.  ...  Each of the 32 examples in a batch was randomly sampled from the full training set with replacement.  ... 
arXiv:1609.08546v2 fatcat:bumbjkkkczgufchjhrsn77whgm

Linking Advanced Visualization and MATLAB for the Analysis of 3D Gene Expression Data [chapter]

Oliver Rübel, Soile V. E. Keränen, Mark Biggin, David W. Knowles, Gunther H. Weber, Hans Hagen, Bernd Hamann, E. Wes Bethel
2012 Mathematics and Visualization  
To maximize the impact of novel, complex data sets, such as PointClouds, the data needs to be accessible to biologists and comprehensible to developers of analysis functions.  ...  Three-dimensional gene expression PointCloud data generated by the Berkeley Drosophila Transcription Network Project (BDTNP) provides quantitative information about the spatial and temporal expression  ...  In addition, PCX provides a convenient GUI for the Matlab functions and parses PointCloud data, i.e., a developer does not need to know how to read PointCloud data.  ... 
doi:10.1007/978-3-642-21608-4_15 fatcat:5bymcaxqgnb3lht6fdne3s5coa

Multi-Resolution Graph Neural Network for Large-Scale Pointcloud Segmentation [article]

Liuyue Xie, Tomotake Furuhata, Kenji Shimada
2020 arXiv   pre-print
While these approaches can relieve the computation burden to some extent, they are still limited in their processing capability for multiple scans.  ...  Dense pointcloud data require a computationally expensive feature encoding process before semantic segmentation.  ...  Intricate details initially present in the dense pointcloud input are removed during the downsampling process. This is undesirable since sparse pointclouds contain less geometric features.  ... 
arXiv:2009.08924v1 fatcat:6sczmd7lzvd3rjh6eevjvlhoza

Combining Implicit Function Learning and Parametric Models for 3D Human Reconstruction [article]

Bharat Lal Bhatnagar, Cristian Sminchisescu, Christian Theobalt, Gerard Pons-Moll
2021 arXiv   pre-print
Implicit functions represented as deep learning approximations are powerful for reconstructing 3D surfaces.  ...  Nevertheless, such features are essential in building flexible models for both computer graphics and computer vision.  ...  Implicit functions represented as deep learning approxima- tions are powerful for reconstructing 3D surfaces.  ... 
arXiv:2007.11432v2 fatcat:mtcxrvdszjc4fhfo7gztc74gqq

AccSS3D: Accelerator for Spatially Sparse 3D DNNs [article]

Om Ji Omer, Prashant Laddha, Gurpreet S Kalsi, Anirud Thyagharajan, Kamlesh R Pillai, Abhimanyu Kulkarni, Anbang Yao, Yurong Chen, Sreenivas Subramoney
2020 arXiv   pre-print
spatially-sparse 3D scene understanding, AccSS3D includes novel spatial locality-aware metadata structures, a near-zero latency and spatial sparsity-aware dataflow optimizer, a surface orientation aware pointcloud  ...  Thus, we correlate SA Avg over randomly picked pointclouds ( Figure 15 ) and observe that: 1.  ...  These components are majorly dominant for the initial and last few layers, where the voxel resolution sizes are higher.  ... 
arXiv:2011.12669v1 fatcat:vhtzlpqvh5ciri2mzmza3cjxqi

Addressing Overfitting on Pointcloud Classification using Atrous XCRF [article]

Hasan Asyari Arief, Ulf Geir Indahl, Geir-Harald Strand, Håvard Tveite
2019 arXiv   pre-print
Advances in techniques for automated classification of pointcloud data introduce great opportunities for many new and existing applications.  ...  Using a GAN training style, we can set up both of the A-XCRF loss functions as the same minimax objective function and perform parameter update using the gradient flow and parameter update in the GAN architecture  ...  Such advantages are important for realtime classification of pointcloud data.  ... 
arXiv:1902.03088v1 fatcat:nboohiqz7fgltkbpzy4rd4g36m

Learning Risk-aware Costmaps for Traversability in Challenging Environments [article]

David D. Fan, Ali-akbar Agha-mohammadi, Evangelos A. Theodorou
2021 arXiv   pre-print
Because we are motivated by preserving the life of the robot, we tackle this learning problem from the perspective of learning tail-risks, i.e. the Conditional Value-at-Risk (CVaR).  ...  ., a probability distribution p(m|x 0:k , z 0:k ) over a possible set M, where z 0:k are sensor observations.  ...  x, α)) outputs, which are fed into the loss function (Equation 13 ).  ... 
arXiv:2107.11722v1 fatcat:lvkbj53vxfcq5bo23zv3vtl6ki

Geometric Affordance Perception: Leveraging Deep 3D Saliency With the Interaction Tensor

Eduardo Ruiz, Walterio Mayol-Cuevas
2020 Frontiers in Neurorobotics  
Our approach works with visually perceived 3D pointclouds and enables to query a 3D scene for locations that support affordances such as sitting or riding, as well as interactions for everyday objects  ...  Specific details on how we train such network are described in the Results and Evaluation section (section 5).  ...  Besides, the manually annotated datasets used for training build on the assumption that objects in the environment have a pre-defined set of affordances; making uncertain how an agent would discover new  ... 
doi:10.3389/fnbot.2020.00045 pmid:32733228 pmcid:PMC7359196 fatcat:gwezipcvgnhuhgm6lpeiq3r5zi
« Previous Showing results 1 — 15 out of 323 results