We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

[신간의 별자리x] 우리/미술, 그리고 ‘슬픔의 박물관’

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

I have developed "tennis elbow" from lugging this book around the past four weeks, but it is worth the pain, the effort, and the aspirin. It is also worth the (relatively speaking) bargain price. Including appendixes, this book contains 894 pages of text. The entire panorama of the neural sciences is surveyed and examined, and it is comprehensive in its scope, from genomes to social behaviors. The editors explicitly state that the book is designed as "an introductory text for students of biology, behavior, and medicine," but it is hard to imagine any audience, interested in any fragment of neuroscience at any level of sophistication, that would not enjoy this book. The editors have done a masterful job of weaving together the biologic, the behavioral, and the clinical sciences into a single tapestry in which everyone from the molecular biologist to the practicing psychiatrist can find and appreciate his or

Principles of Neural Science

Despite the breakthroughs in accuracy and speed of single image super-resolution using faster and deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover the finer texture details when we super-resolve at large upscaling factors? The behavior of optimization-based super-resolution methods is principally driven by the choice of the objective function. Recent work has largely focused on minimizing the mean squared reconstruction error. The resulting estimates have high peak signal-to-noise ratios, but they are often lacking high-frequency details and are perceptually unsatisfying in the sense that they fail to match the fidelity expected at the higher resolution. In this paper, we present SRGAN, a generative adversarial network (GAN) for image super-resolution (SR). To our knowledge, it is the first framework capable of inferring photo-realistic natural images for 4x upscaling factors. To achieve this, we propose a perceptual loss function which consists of an adversarial loss and a content loss. The adversarial loss pushes our solution to the natural image manifold using a discriminator network that is trained to differentiate between the super-resolved images and original photo-realistic images. In addition, we use a content loss motivated by perceptual similarity instead of similarity in pixel space. Our deep residual network is able to recover photo-realistic textures from heavily downsampled images on public benchmarks. An extensive mean-opinion-score (MOS) test shows hugely significant gains in perceptual quality using SRGAN. The MOS scores obtained with SRGAN are closer to those of the original high-resolution images than to those obtained with any state-of-the-art method.

/pdf/photo-realistic-single-image-super-resolution-using-a-2y0438rlmd.pdf

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

The invention discloses a method for acquiring three-dimensional information of frequency mixing structured light based on phase encoding, mainly solving the problems of low measuring precision, low spatial resolution and long consumed time of the traditional three-dimensional information acquiring method. The method is realized through the steps: designing a double-color stripe template with two kinds of frequency information and changed strength; projecting the double-color stripe template on an objected to be measured by using a projector, and recording a deformed stripe image by using a camera; solving the color and strength phase distribution of the deformed stripe image; calculating truncation phase expansion values of pixel points in the deformed stripe image according to the color and strength phase distribution; confirming matching points of the pixel points in the projected template according to the truncation phase expansion values; and solving the three-dimensional coordinate value of each pixel point in the deformed stripe image according to a triangulation theory and the coordinates of the matching points. The method has the advantages of high space resolution, high measuring precision and high measuring speed and can be used for acquiring the three-dimensional information of dynamic objects.

Method for acquiring three-dimensional information of frequency mixing structured light based on phase encoding

In this paper, we propose a new adaptive lifting scheme that not only locally adapts the filtering directions to the orientations of image features, but also adapts the lifting filters to the statistic properties of image signal. The proposed approach refines previous adaptive directional lifting-based wavelet transform (ADL) by combining directional lifting and adaptive lifting filters to form a unified framework. The image signal is first segmented into regions of textures and edges with close directional features. The lifting filters are then effectively designed. The prediction step is designed to minimize the prediction error of the image signal, and the update step is designed to minimize the reconstruction error. Significant improvements on objective and subjective quality over conventional 2-D wavelet transform and previous ADL transform are achieved.

Signal-adapted directional lifting scheme for image compression

A new multiple description coding (MDC) approach is proposed based on the theory of compressive sensing (CS). The CS theory allows a signal to be reconstructed from a small number of its random measurements if the signal is sparse in some space. An attractive property of CS for MDC applications is that the reconstruction error only depends on the number but not on which of the transmitted measurements that are received. By treating each CS measurement as a description, we have a balanced MDC scheme with fine description granularity and low encoding complexity. Another advantage of the new MDC approach is that all signals can be coded the same but decoded in different spaces for better sparse reconstruction.

A compressive sensing approach of multiple descriptions for network multimedia communication

Sparse representation (SR)-based SAR imaging approaches have shown their superior performance compared with conventional approaches. However, for an image with rich spatial structures, a fixed global dictionary is usually ineffective to characterize the local structures. Piecewise autoregressive (PAR) model indicates that each pixel can be linearly represented by its local neighboring pixels. Inspired by this, an adaptive sparse space, effectively characterizing the varying image local structures, is designed, in which the entries are derived from the PAR model. By incorporating the adaptive SR into the SAR imaging, a novel structural SR-based SAR (SSR-SAR) imaging approach is proposed. Due to the fact that the adaptive sparse space is greatly dependent on the prior information of the SAR image, updating of the adaptive sparse space and SAR imaging is a joint optimization problem. In our approach, we propose to introduce the alternative minimization scheme to solve the problem. Besides, the Augmented Lagrangian Multiplier technique is adopted to accelerate the computation speed. Finally, experimental results are shown to demonstrate the validity of the proposed approach.

SAR Imaging With Structural Sparse Representation

This correspondence proposes a novel method for designing directional filter banks (DFBs) with arbitrary number of subbands Its key feature is the ability to decompose images into arbitrary directionally oriented subbands The proposed approach is based on the pseudo-polar Fourier transform (PPFT) and one-dimensional (1-D) filter banks (FBs) We take some modifications on the PPFT and then employ 1-D FBs to the modified PPFT With these operations, the two-dimensional (2-D) DFBs are obtained and the design of them is converted to that of 1-D FBs plus a modified PPFT Since the number of channels of 1-D FBs can be arbitrary, the 2-D DFBs with arbitrary number of subbands can be achieved which is highly expected for directional representations of images Two examples on directional feature extractions illustrate that the proposed non-2n channel DFBs can capture directional information more flexibly than the existing methods

Guangming Shi

Papers

Method for acquiring three-dimensional information of frequency mixing structured light based on phase encoding

Signal-adapted directional lifting scheme for image compression

A compressive sensing approach of multiple descriptions for network multimedia communication

SAR Imaging With Structural Sparse Representation

Design of Directional Filter Banks With Arbitrary Number of Subbands