Professional Documents
Culture Documents
Download pdf Advances In Neural Computation Machine Learning And Cognitive Research Iii Selected Papers From The Xxi International Conference On Neuroinformatics October 7 11 2019 Dolgoprudny Moscow Region Russia ebook full chapter
Download pdf Advances In Neural Computation Machine Learning And Cognitive Research Iii Selected Papers From The Xxi International Conference On Neuroinformatics October 7 11 2019 Dolgoprudny Moscow Region Russia ebook full chapter
Boris Kryzhanovsky
Witali Dunin-Barkowski
Vladimir Redko
Yury Tiumentsev Editors
Advances in Neural
Computation, Machine
Learning, and
Cognitive Research III
Selected Papers from the XXI
International Conference on
Neuroinformatics, October 7–11, 2019,
Dolgoprudny, Moscow Region, Russia
Studies in Computational Intelligence
Volume 856
Series Editor
Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
The series “Studies in Computational Intelligence” (SCI) publishes new develop-
ments and advances in the various areas of computational intelligence—quickly and
with a high quality. The intent is to cover the theory, applications, and design
methods of computational intelligence, as embedded in the fields of engineering,
computer science, physics and life sciences, as well as the methodologies behind
them. The series contains monographs, lecture notes and edited volumes in
computational intelligence spanning the areas of neural networks, connectionist
systems, genetic algorithms, evolutionary computation, artificial intelligence,
cellular automata, self-organizing systems, soft computing, fuzzy systems, and
hybrid intelligent systems. Of particular value to both the contributors and the
readership are the short publication timeframe and the world-wide distribution,
which enable both wide and rapid dissemination of research output.
The books of this series are submitted to indexing to Web of Science,
EI-Compendex, DBLP, SCOPUS, Google Scholar and Springerlink.
Editors
Advances in Neural
Computation, Machine
Learning, and Cognitive
Research III
Selected Papers from the XXI International
Conference on Neuroinformatics,
October 7–11, 2019, Dolgoprudny,
Moscow Region, Russia
123
Editors
Boris Kryzhanovsky Witali Dunin-Barkowski
Scientific Research Institute for System Scientific Research Institute for System
Analysis of Russian Academy of Sciences Analysis of Russian Academy of Sciences
Moscow, Russia Moscow, Russia
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
Boris Kryzhanovsky
Witali Dunin-Barkowski
Vladimir Redko
Yury Tiumentsev
v
Organization
Editorial Board
Boris Kryzhanovsky Scientific Research Institute for System Analysis
of Russian Academy of Sciences
Witali Dunin-Barkowsky Scientific Research Institute for System Analysis
of Russian Academy of Sciences
Vladimir Red’ko Scientific Research Institute for System Analysis
of Russian Academy of Sciences
Yury Tiumentsev Moscow Aviation Institute
(National Research University)
Advisory Board
vii
viii Organization
Physical Address:
KEDRI
Auckland University of Technology
AUT Tower, Level 7
Corner Rutland and Wakefield Street
Auckland
Postal Address:
KEDRI
Auckland University of Technology
Private Bag 92006
Auckland 1142
New Zealand
Prof. Jun Wang, PhD, FIEEE, FIAPR
Chair Professor of Computational Intelligence
Department of Computer Science
City University of Hong Kong
Kowloon Tong, Kowloon, Hong Kong
+852 34429701 (tel.)
+852-34420503 (fax)
jwang.cs@cityu.edu.hk
Co-chairs
Program Committee
Invited Papers
Deep Learning a Single Photo Voxel Model Prediction from Real
and Synthetic Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Vladimir V. Kniaz, Peter V. Moshkantsev, and Vladimir A. Mizginov
Tensor Train Neural Networks in Retail Operations . . . . . . . . . . . . . . . 17
Serge A. Terekhov
Semi-empirical Neural Network Based Modeling and Identification
of Controlled Dynamical Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Yury Tiumentsev and Mikhail Egorchev
Artificial Intelligence
Photovoltaic System Control Model on the Basis of a Modified
Fuzzy Neural Net . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Ekaterina A. Engel and Nikita E. Engel
Impact of Assistive Control on Operator Behavior Under High
Operational Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Mikhail Kopeliovich, Evgeny Kozubenko, Mikhail Kashcheev,
Dmitry Shaposhnikov, and Mikhail Petrushan
Hierarchical Actor-Critic with Hindsight for Mobile Robot
with Continuous State Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Staroverov Aleksey and Aleksandr I. Panov
The Hybrid Intelligent Information System for Music Classification . . . 71
Aleksandr Stikharnyi, Alexey Orekhov, Ark Andreev,
and Yuriy Gapanyuk
The Hybrid Intelligent Information System for Poems Generation . . . . 78
Maria Taran, Georgiy Revunkov, and Yuriy Gapanyuk
xiii
xiv Contents
Deep Learning
The Simple Approach to Multi-label Image Classification Using
Transfer Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Yuriy S. Fedorenko
Application of Deep Neural Network for the Vision System
of Mobile Service Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Nikolay Filatov, Vladislav Vlasenko, Ivan Fomin,
and Aleksandr Bakhshiev
Research on Convolutional Neural Network for Object
Classification in Outdoor Video Surveillance System . . . . . . . . . . . . . . . 221
I. S. Fomin and A. V. Bakhshiev
Post-training Quantization of Deep Neural Network Weights . . . . . . . . 230
E. M. Khayrov, M. Yu. Malsagov, and I. M. Karandashev
Deep-Learning Approach for McIntosh-Based Classification
Of Solar Active Regions Using HMI and MDI Images . . . . . . . . . . . . . . 239
Irina Knyazeva, Andrey Rybintsev, Timur Ohinko,
and Nikolay Makarenko
Deep Learning for ECG Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . 246
Viktor Moskalenko, Nikolai Zolotykh, and Grigory Osipov
Competitive Maximization of Neuronal Activity in Convolutional
Recurrent Spiking Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
Dmitry Nekhaev and Vyacheslav Demin
A Method of Choosing a Pre-trained Convolutional Neural
Network for Transfer Learning in Image Classification Problems . . . . . 263
Alexander G. Trofimov and Anastasia A. Bogatyreva
The Usage of Grayscale or Color Images for Facial Expression
Recognition with Deep Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . 271
Dmitry A. Yudin, Alexandr V. Dolzhenko, and Ekaterina O. Kapustina
xvi Contents
1 Introduction
elements that represent slices of the camera’s frustum. Each fruxel is aligned with
the pixel of the input color image (see Fig. 1). Fruxel models facilitate robust
training of a view-centered model as the contour alignment between the input
image and the fruxel model is preserved.
To the best of our knowledge, there are no results in the literature regarding
view-centered voxel model dataset generation using synthetic images and 3D
modeling. In this paper, we explore the impact of the synthetic data in the
performance of a view-centered model. We use a recently proposed generative
adversarial model Z-GAN [24] as a starting point for our research. We prepared
an extensive SyntheticVoxels dataset with 2k synthetic images of three object
classes and corresponding ground truth fruxel models. We made our dataset
publicly available. We compare the performance of the Z-GAN model trained on
real, synthetic, and mixed data.
The results of joint training on the synthetic and real data are encouraging
and show that synthetic data allows the model to generalize to previously unseen
objects. The developed view-centered dataset generation technique allows mod-
eling challenging 3D object configurations and traffic situations that can not be
reconstructed online using laser scanning or similar approaches.
2 Related Work
Generative Adversarial Networks. Development of a new type of neural
networks known as Generative Adversarial Networks (GANs) [14] made it possi-
ble to provide a mapping from a random noise vector to a domain of the desired
outputs (e.g., images, voxel models). GANs have received a lot of scholar atten-
tion in recent years. These networks provide inspiring results in such tasks as
image-to-image translation [20] and the voxel model generation [42].
Methods that leverage a latent space for 3D shape synthesis were developed
recently [5,13,42]. Wu et al. have proposed a GAN model [42] for a voxel model
generation (3D-GAN). This made it possible to predict models with a resolution
64 × 64 × 64 elements from a randomly sampled noise vector. The developed
method was used for a single-image 3D reconstruction using an approach pro-
posed in [13]. Despite the fact that 3D-GAN increased the number of elements in
the model compared to [13], the generalization ability of this method was low,
especially for previously unseen objects.
3 Method
The aim of the present research is to compare the performance of a single photo
voxel model prediction method trained on synthetic, real and mixed data. In
our research we use a generative adversarial network Z-GAN [24] that performs
color image-to-voxel model translation. Z-GAN model uses a special kind of voxel
model in which the voxel model is aligned with an input image.
While a depth map that present distances only to the object surface from
a given viewpoint, the voxel model includes information about the entire 3D
scene. The proposed frustum voxel models combines features of a depth map
and a voxel model. We use a hypothesis made by [41] as the starting point
for our research. To provide the aligned voxel model, we combine depth map
representation with a voxel grid. We term the resulting 3D model as a Frustum
Voxel model (Fruxel model).
deconv3D
deconv3D
deconv3D
deconv3D
deconv3D
deconv3D
deconv3D
deconv3D
4×4×4
4×4×4
4×4×4
4×4×4
4×4×4
2×4×4
4×4×4
2×4×4
1 × 1 × 1 × 1024
2 × 2 × 2 × 1024
4 × 4 × 4 × 1024
16 × 16 × 16 × 1024 8 × 8 × 8 × 1024
128 × 128 × 128 128 × 128 × 128 × 128 64 × 64 × 64 × 256 32 × 32 × 32 × 512
We use pix2pix [20] framework as a base to develop our Z-GAN model. We keep
the encoder part of the generator unchanged. We change 2D convolution layers
with 3D deconvolution layers to encode a correlation between neighbor slices
along the Z-axis.
We keep the skip connections between the layers of the same depth that
were proposed in the U-Net model [34]. We believe that skip connections help
to transfer high-frequency components of the input image to the high-frequency
components of the 3D shape.
Fig. 3. Synthetic dataset generation technique: (a) virtual camera, (b) slice of
fruxel model, (c) cutting plane, (d) low-poly 3D model, (e) synthetic color image.
We use d copies of each channel of F2D to fill the third dimension of F3D . We
term this operation as “copy inflate”. The architecture of generator is presented
in Fig. 2.
Fig. 4. Examples of color images and corresponding fruxel models from our Synthet-
icVoxels dataset. Fruxel models are presented as depth maps in pseudo-colors.
cutting plane movement using the Blender Python API. We use an additional
ground plane to provide realistic object shadows. We render the plane with shad-
ows separately and use alpha-compositing to obtain the final synthetic image.
SyntheticVoxels Dataset. Examples of synthetic images with ground truth
fruxel models from our SyntheticVoxels dataset are presented in Figs. 4 and 5.
The dataset includes images and fruxel models for four object classes: car, truck,
off-road vehicle, and van.
4 Experiments
4.1 Network Training
Our Z-GAN framework was trained on the VoxelCity [24] and SyntheticVoxels
datasets using PyTorch library [29]. We use independent test splits of Synthet-
icVoxels and VoxelCity datasets for evaluation with fruxel model parameters
{zn = 2, zf = 12, d = 128, α = 40◦ }. The training was performed using the
NVIDIA 1080 Ti GPU and took 20 hours for the whole framework. For network
optimization, we use a minibatch stochastic gradient descent with an Adam
solver. We set the learning rate to 0.0002 with momentum parameters β1 = 0.5,
β2 = 0.999 similar to [20].
Car
Van
Fig. 5. Examples of color images and corresponding fruxel models from our Synthet-
icVoxels dataset. Fruxel models are presented as depth maps in pseudo-colors.
Fig. 7. Qualitative evaluation on real images from VoxelCity dataset. Fruxel models
are presented as depth maps in pseudo-colors.
Hamiltonian Mechanics 13
Table 1. IoU metric for different object classes for Z-GAN model trained on real, syn-
thetic and mixed data.
5 Conclusions
Acknowledgments. The reported study was funded by Russian Foundation for Basic
Research (RFBR) according to the project No 17-29-04410, and by the Russian Science
Foundation (RSF) according to the research project No 19-11-11008.
14 V. V. Kniaz et al.
References
1. Balntas, V., Doumanoglou, A., Sahin, C., Sock, J., Kouskouridas, R., Kim, T.:
Pose guided RGBD feature learning for 3d object pose estimation. In: IEEE Inter-
national Conference on Computer Vision, ICCV 2017, Venice, Italy, 22–29 October
2017, pp. 3876–3884 (2017). https://doi.org/10.1109/ICCV.2017.416
2. Balntas, V., Doumanoglou, A., Sahin, C., Sock, J., Kouskouridas, R., Kim, T.K.:
Pose guided RGBD feature learning for 3D object pose estimation. In: The IEEE
International Conference on Computer Vision (ICCV) (2017)
3. Brachmann, E., Krull, A., Nowozin, S., Shotton, J., Michel, F., Gumhold, S.,
Rother, C.: DSAC - differentiable RANSAC for camera localization. In: The IEEE
Conference on Computer Vision and Pattern Recognition (CVPR) (2017)
4. Brachmann, E., Rother, C.: Learning less is more - 6d camera localization via
3d surface regression. In: The IEEE Conference on Computer Vision and Pattern
Recognition (CVPR) (2018)
5. Brock, A., Lim, T., Ritchie, J., Weston, N.: Generative and discriminative voxel
modeling with convolutional neural networks, pp. 1–9 (2016). https://nips.cc/
Conferences/2016. Workshop contribution; Neural Information Processing Con-
ference : 3D Deep Learning, NIPS, 05–12 Dec 2016
6. Chang, A.X., Funkhouser, T.A., Guibas, L.J., Hanrahan, P., Huang, Q.X., Li, Z.,
Savarese, S., Savva, M., Song, S., Su, H., Xiao, J., Yi, L., Yu, F.: Shapenet: an
information-rich 3d model repository (2015). CoRR arXiv:abs/1512.03012
7. Choy, C.B., Xu, D., Gwak, J., Chen, K., Savarese, S.: 3d-r2n2: a unified approach
for single and multi-view 3d object reconstruction. In: Proceedings of the European
Conference on Computer Vision (ECCV) (2016)
8. Doumanoglou, A., Kouskouridas, R., Malassiotis, S., Kim, T.: Recovering 6d object
pose and predicting next-best-view in the crowd. In: 2016 IEEE Conference on
Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV, USA,
27–30 June 2016, pp. 3583–3592 (2016). https://doi.org/10.1109/CVPR.2016.390
9. Drost, B., Ulrich, M., Bergmann, P., Hartinger, P., Steger, C.: Introducing mvtec
itodd - a dataset for 3d object recognition in industry. In: The IEEE International
Conference on Computer Vision (ICCV) Workshops (2017)
10. El-Hakim, S.: A flexible approach to 3d reconstruction from single images. In: ACM
SIGGRAPH, vol. 1, pp. 12–17 (2001)
11. Everingham, M., Van Gool, L., Williams, C.K.I., Winn, J., Zisserman, A.: The
pascal visual object classes (VOC) challenge. Int. J. Comput. Vis. 88(2), 303–338
(2009)
12. Firman, M., Mac Aodha, O., Julier, S., Brostow, G.J.: Structured prediction of
unobserved voxels from a single depth image. In: The IEEE Conference on Com-
puter Vision and Pattern Recognition (CVPR) (2016)
13. Girdhar, R., Fouhey, D.F., Rodriguez, M., Gupta, A.: Learning a predictable and
generative vector representation for objects, chap. 34, pp. 702–722. Springer, Cham
(2016)
14. Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair,
S., Courville, A., Bengio, Y.: Generative adversarial nets. In: Advances in Neural
Information Processing Systems, pp. 2672–2680 (2014)
15. Hinterstoisser, S., Lepetit, V., Ilic, S., Holzer, S., Bradski, G., Konolige, K., Navab,
N.: Model based training, detection and pose estimation of texture-less 3d objects
in heavily cluttered scenes. In: Asian Conference on Computer Vision, pp. 548–562.
Springer, Heidelberg (2012)
Another random document with
no related content on Scribd:
The Project Gutenberg eBook of Less than kin
This ebook is for the use of anyone anywhere in the United
States and most other parts of the world at no cost and with
almost no restrictions whatsoever. You may copy it, give it away
or re-use it under the terms of the Project Gutenberg License
included with this ebook or online at www.gutenberg.org. If you
are not located in the United States, you will have to check the
laws of the country where you are located before using this
eBook.
Language: English
New York
Henry Holt and Company
1909
Copyright, 1909
BY
HENRY HOLT AND COMPANY