检索结果-内蒙古大学图书馆

Learning pose specific representations by predicting different views

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Poier, Georg Schinagl, David Bischof, Horst Institute for Computer Graphics and Vision Graz University of Technology Austria

The labeled data required to learn pose estimation for articulated objects is difficult to provide in the desired quantity, realism, density, and accuracy. To address this issue, we develop a method to learn representations, which are very specific for articulated poses, without the need for labeled training data. We exploit the observation that the object pose of a known object is predictive for the appearance in any known view. That is, given only the pose and shape parameters of a hand, the hand's appearance from any viewpoint can be approximated. To exploit this observation, we train a model that - given input from one view - estimates a latent representation, which is trained to be predictive for the appearance of the object when captured from another viewpoint. Thus, the only necessary supervision is the second view. The training process of this model reveals an implicit pose representation in the latent space. Importantly, at test time the pose representation can be inferred using only a single view. In qualitative and quantitative experiments we show that the learned representations capture detailed pose information. Moreover, when training the proposed method jointly with labeled and unlabeled data, it consistently surpasses the performance of its fully supervised counterpart, while reducing the amount of needed labeled samples by at least one order of magnitude. Copyright © 2018, The Authors. All rights reserved.

关键词： Machine learning

MURAUER: Mapping Unlabeled Real Data for Label AUstERity

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Poier, Georg Opitz, Michael Schinagl, David Bischof, Horst Institute for Computer Graphics and Vision Graz University of Technology Austria

Data labeling for learning 3D hand pose estimation models is a huge effort. Readily available, accurately labeled synthetic data has the potential to reduce the effort. However, to successfully exploit synthetic data, current state-of-the-art methods still require a large amount of labeled real data. In this work, we remove this requirement by learning to map from the features of real data to the features of synthetic data mainly using a large amount of synthetic and unlabeled real data. We exploit unlabeled data using two auxiliary objectives, which enforce that (i) the mapped representation is pose specific and (ii) at the same time, the distributions of real and synthetic data are aligned. While pose specifity is enforced by a self-supervisory signal requiring that the representation is predictive for the appearance from different views, distributions are aligned by an adversarial term. In this way, we can significantly improve the results of the baseline system, which does not use unlabeled data and outperform many recent approaches already with about 1% of the labeled real data. This presents a step towards faster deployment of learning based hand pose estimation, making it accessible for a larger range of applications. Copyright © 2018, The Authors. All rights reserved.

关键词： Machine learning

The quest for the golden activation function

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Basirat, Mina Roth, Peter M. Graz University of Technology Institute of Computer Graphics and Vision Austria

Deep Neural Networks have been shown to be beneficial for a variety of tasks, in particular allowing for end-to-end learning and reducing the requirement for manual design decisions. However, still many parameters have to be chosen in advance, also raising the need to optimize them. One important, but often ignored system parameter is the selection of a proper activation function. Thus, in this paper we target to demonstrate the importance of activation functions in general and show that for different tasks different activation functions might be meaningful. To avoid the manual design or selection of activation functions, we build on the idea of genetic algorithms to learn the best activation function for a given task. In addition, we introduce two new activation functions, ELiSH and HardELiSH, which can easily be incorporated in our framework. In this way, we demonstrate for three different image classification benchmarks that different activation functions are learned, also showing improved results compared to typically used baselines. Copyright © 2018, The Authors. All rights reserved.

关键词： Deep neural networks

Variational networks for joint image reconstruction and classification of tumor immune cell interactions in melanoma tissue sections

学校读者我要写书评

暂无评论

Variational networks for joint image reconstruction and clas...

Workshop on Bildverarbeitung fur die Medizin, 2018

作者： Effland, Alexander Hölzel, Michael Klatzer, Teresa Kobler, Erich Landsberg, Jennifer Neuhäuser, Leonie Pock, Thomas Rumpf, Martin Institute for Numerical Simulation University of Bonn Bonn Germany Institute of Clinical Chemistry and Clinical Pharmacology University of Bonn Bonn Germany Institute of Computer Graphics and Vision Graz University of Technology Graz Austria Department of Dermatology and Allergy University of Bonn Bonn Germany

ISBN: (纸本)9783662580950

Immunotherapy is currently revolutionizing the treatment of cancer. Detailed analyses of tumor immune cell interaction in the tumor microenvironment will facilitate an accurate prediction of a patient’s clinical response. The automatic and reliable pre-screening of histological tissue sections for tumor infiltrating immune cells (TILs) will support the development of TIL-based predictive biomarkers for checkpoint immunotherapy. In this paper, a learning approach for image classification is presented, which allows various pattern inquires for different types of tissue section images. The underlying trainable reaction diffusion model combines classification and denoising. The model is trained using a stochastic generation of training data. The effectiveness of this approach is demonstrated for immunofluorescent and for Hematoxylin and Eosin (H&E) stained melanoma section images. A particular focus is on the classification of TILs in the proximity to melanoma cells in an experimental melanoma mouse model and in human melanoma. This new learning approach for images of melanoma tissue sections will refine the strategy for the practical clinical application of biomarker research. © Springer-Verlag GmbH Deutschland 2018.

关键词： Tumors

Deep metric learning with BIER: Boosting independent embeddings robustly

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Opitz, Michael Waltner, Georg Possegger, Horst Bischof, Horst Institute of Computer Graphics and Vision Graz University of Technology Graz8010 Austria

Learning similarity functions between image pairs with deep neural networks yields highly correlated activations of embeddings. In this work, we show how to improve the robustness of such embeddings by exploiting the independence within ensembles. To this end, we divide the last embedding layer of a deep network into an embedding ensemble and formulate training this ensemble as an online gradient boosting problem. Each learner receives a reweighted training sample from the previous learners. Further, we propose two loss functions which increase the diversity in our ensemble. These loss functions can be applied either for weight initialization or during training. Together, our contributions leverage large embedding sizes more effectively by significantly reducing correlation of the embedding and consequently increase retrieval accuracy of the embedding. Our method works with any differentiable loss function and does not introduce any additional parameters during test time. We evaluate our metric learning method on image retrieval tasks and show that it improves over state-of-the-art methods on the CUB-200-2011, Cars-196, Stanford Online Products, In-Shop Clothes Retrieval and VehicleID datasets. Copyright © 2018, The Authors. All rights reserved.

关键词： Embeddings

SHREC'19 track: Extended 2D scene sketch-based 3D scene retrieval 12

学校读者我要写书评

暂无评论

SHREC'19 track: Extended 2D scene sketch-based 3D scene retr...

12th Eurographics Workshop on 3D Object Retrieval, 3DOR 2019, in conjunction with the 40th Annual Conference of the European Association for computer graphics, EG 2019

作者： Yuan, Juefei Abdul-Rashid, Hameed Li, Bo Lu, Yijuan Schreck, Tobias Bui, Ngoc-Minh Do, Trong-Le Nguyen, Khac-Tuan Nguyen, Thanh-An Nguyen, Vinh-Tiep Tran, Minh-Triet Wang, Tianyang School of Computing Sciences and Computer Engineering University of Southern Mississippi United States Department of Computer Science Texas State University United States Institute of Computer Graphics and Knowledge Visualization Graz University of Technology Austria Faculty of Information Technology Vietnam National University Ho Chi Minh City Viet Nam Software Engineering Lab Vietnam National University Ho Chi Minh City Viet Nam University of Information Technology Vietnam National University Ho Chi Minh City Viet Nam Department of Computer Science and Information Technology Austin Peay State University United States

ISBN: (纸本)9783038680772

Sketch-based 3D scene retrieval is to retrieve 3D scene models given a user's hand-drawn 2D scene sketch. It is a brand new but also very challenging research topic in the field of 3D object retrieval due to the semantic gap in their representations: 3D scene models or views differ from non-realistic 2D scene sketches. To boost this interesting research, we organized a 2D Scene Sketch-Based 3D Scene Retrieval track in SHREC'18, resulting a SceneSBR18 benchmark which contains 10 scene classes. In order to make it more comprehensive, we have extended the number of the scene categories from the initial 10 classes in the SceneSBR2018 benchmark to 30 classes, resulting in a new and more challenging benchmark SceneSBR2019 which has 750 2D scene sketches and 3,000 3D scene models. Therefore, the objective of this track is to further evaluate the performance and scalability of different 2D scene sketch-based 3D scene model retrieval algorithms using this extended and more comprehensive new benchmark. In this track, two groups from USA and Vietnam have successfully submitted 4 runs. Based on 7 commonly used retrieval metrics, we evaluate their retrieval performance. We have also conducted a comprehensive analysis and discussion of these methods and proposed several future research directions to deal with this challenging research topic. Deep learning techniques have been proved their great potentials again in dealing with this challenging retrieval task, in terms of both retrieval accuracy and scalability to a larger dataset. We hope this publicly available benchmark, together with its evaluation results and source code, will further enrich and promote 2D scene sketch-based 3D scene retrieval research area and its corresponding applications. © 2019 The Author(s)

关键词： Scalability

Deep 2.5D vehicle classification with sparse SfM depth prior for automated toll systems

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Waltner, Georg Maurer, Michael Holzmann, Thomas Ruprecht, Patrick Opitz, Michael Possegger, Horst Fraundorfer, Friedrich Bischof, Horst Institute of Computer Graphics and Vision Faculty of Computer Science Graz University of Technology Graz8010 Austria

Automated toll systems rely on proper classification of the passing vehicles. This is especially difficult when the images used for classification only cover parts of the vehicle. To obtain information about the whole vehicle. we reconstruct the vehicle as 3D object and exploit this additional information within a Convolutional Neural Network (CNN). However, when using deep networks for 3D object classification, large amounts of dense 3D models are required for good accuracy, which are often neither available nor feasible to process due to memory requirements. Therefore, in our method we reproject the 3D object onto the image plane using the reconstructed points, lines or both. We utilize this sparse depth prior within an auxiliary network branch that acts as a regularizer during training. We show that this auxiliary regularizer helps to improve accuracy compared to 2D classification on a real-world dataset. Furthermore due to the design of the network, at test time only the 2D camera images are required for classification which enables the usage in portable computer vision systems. Copyright © 2018, The Authors. All rights reserved.

关键词： Classification (of information)

CG-SENSE revisited: Results from the first ISMRM reproducibility challenge

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Maier, Oliver Baete, Steven H. Fyrdahl, Alexander Hammernik, Kerstin Harrevelt, Seb Kasper, Lars Karakuzu, Agah Loecher, Michael Patzig, Franz Tian, Ye Wang, Ke Gallichan, Daniel Uecker, Martin Knoll, Florian Institute of Medical Engineering Graz University of Technology Graz Austria Center for Biomedical Imaging New York University School of Medicine New YorkNY United States Department of Clinical Physiology Karolinska University Hospital Karolinska Institutet Stockholm Sweden Department of Computing Imperial College London London United Kingdom Institute of Computer Graphics and Vision Graz University of Technology Graz Austria Department of Biomedical Engineering Eindhoven University of Technology Eindhoven Netherlands Institute for Biomedical Engineering ETH Zurich University of Zurich Zurich Switzerland Translational Neuromodeling Unit Institute for Biomedical Engineering University of Zurich ETH Zurich Zurich Switzerland NeuroPoly Lab Institute of Biomedical Engineering Polytechnique Montréal Montréal Canada Department of Radiology Stanford University StanfordCA United States Department of Radiology and Imaging Sciences University of Utah Salt Lake CityUT United States Ming Hsieh Department of Electrical and Computer Engineering Viterbi School of Engineering University of Southern California Los AngelesCA United States Department of Electrical Engineering and Computer Sciences University of California Berkeley BerkeleyCA United States Cardiff University Brain Research Imaging Centre Cardiff United Kingdom Institute for Diagnostic and Interventional Radiology University Medical Center Göttingen Göttingen Germany Germany University of Göttingen Germany University of Göttingen Germany

Purpose The aim of this work is to shed light on the issue of reproducibility in MR image reconstruction in the context of a challenge. Participants had to recreate the results of "Advances in sensitivity encoding with arbitrary k-space trajectories" by Pruessmann et al. Methods The task of the challenge was to reconstruct radially acquired multi-coil k-space data (brain/heart) following the method in the original paper, reproducing its key figures. Results were compared to consolidated reference implementations created after the challenge, accounting for the two most common programming languages used in the submissions (Matlab/Python). Results Visually, differences between submissions were small. Pixel-wise differences originated from image orientation, assumed field-of-view or resolution. The reference implementations were in good agreement, both visually and in terms of image similarity metrics. Discussion and Conclusion While the description level of the published algorithm enabled participants to reproduce CG-SENSE in general, details of the implementation varied, e.g., density compensation or Tikhonov regularization. Implicit assumptions about the data lead to further differences, emphasizing the importance of sufficient meta-data accompanying open data sets. Defining reproducibility quantitatively turned out to be non-trivial for this image reconstruction challenge, in the absence of ground-truth results. Typical similarity measures like NMSE of SSIM were misled by image intensity scaling and outlier pixels. Thus, to facilitate reproducibility, researchers are encouraged to publish code and data alongside the original paper. Future methodological papers on MR image reconstruction might benefit from the consolidated reference implementations of CG-SENSE presented here, as a benchmark for methods comparison. Copyright © 2020, The Authors. All rights reserved.

关键词： Magnetic resonance imaging

Visual Analysis of Urban Traffic Data based on High-Resolution and High-Dimensional Environmental Sensor Data 6

学校读者我要写书评

暂无评论

Visual Analysis of Urban Traffic Data based on High-Resoluti...

6th Workshop on Visualisation in Environmental Sciences, EnvirVis 2018 at EuroVis 2018

作者： Häußler, Johannes Stein, Manuel Seebacher, Daniel Janetzko, Halldor Schreck, Tobias Keim, Daniel A. University of Konstanz Department of Computer and Information Science Germany University of Zurich Department of Geography Switzerland Graz University of Technology Institute for Computer Graphics and Knowledge Visualization Austria

ISBN: (纸本)9783038680635

Urbanization is an increasing global trend resulting in a strong increase in public and individual transportation needs. Accordingly, a major challenge for traffic and urban planners is the design of sustainable mobility concepts to maintain and increase the long-term health of humans by reducing environmental pollution. Recent developments in sensor technology allow the precise tracking of vehicle sensor information, allowing a closer and more in-depth analysis of traffic data. We propose a visual analytics system for the exploration of environmental factors in these high-resolution and high-dimensional mobility sensor data. Additionally, we introduce an interactive visual logging approach to enable experts to cope with complex interactive analysis processes and the problem of the reproducibility of results. The usefulness of our approach is demonstrated via two expert studies with two domain experts from the field of environment-related projects and urban traffic planning. © 2018 The Author(s) Eurographics Proceedings © 2018 The Eurographics Association.

关键词： computer graphics