TensorFlow is an interface for expressing machine learning algorithms, and an
implementation for executing such algorithms. A computation expressed using
TensorFlow can be executed with little or no change on a wide variety of
heterogeneous systems, ranging from mobile devices such as phones and tablets
up to large-scale distributed systems of hundreds of machines and thousands of
computational devices such as GPU cards. The system is flexible and can be used
to express a wide variety of algorithms, including training and inference
algorithms for deep neural network models, and it has been used for conducting
research and for deploying machine learning systems into production across more
than a dozen areas of computer science and other fields, including speech
recognition, computer vision, robotics, information retrieval, natural language
processing, geographic information extraction, and computational drug
discovery. This paper describes the TensorFlow interface and an implementation
of that interface that we have built at Google. The TensorFlow API and a
reference implementation were released as an open-source package under the
Apache 2.0 license in November, 2015 and are available at www.tensorflow.org.

/pdf/tensorflow-large-scale-machine-learning-on-heterogeneous-2yk4fgchw6.pdf

TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems

In this paper, a new technique for modeling textured 3D faces is introduced. 3D faces can either be generated automatically from one or more photographs, or modeled directly through an intuitive user interface. Users are assisted in two key problems of computer aided face modeling. First, new face images or new 3D face models can be registered automatically by computing dense one-to-one correspondence to an internal face model. Second, the approach regulates the naturalness of modeled faces avoiding faces with an “unlikely” appearance. Starting from an example set of 3D face models, we derive a morphable face model by transforming the shape and texture of the examples into a vector space representation. New faces and expressions can be modeled by forming linear combinations of the prototypes. Shape and texture constraints derived from the statistics of our example faces are used to guide manual modeling or automated matching algorithms. We show 3D face reconstructions from single images and their applications for photo-realistic image manipulations. We also demonstrate face manipulations according to complex parameters such as gender, fullness of a face or its distinctiveness.

/pdf/a-morphable-model-for-the-synthesis-of-3d-faces-1fz06uy6kp.pdf

A morphable model for the synthesis of 3D faces

There is a special reason for reviewing this book at this time: it is the 50th edition of a compendium that is known and used frequently in most chemical and physical laboratories in many parts of the world. Surely, a publication that has been published for 56 years, withstanding the vagaries of science in this century, must have had something to offer. There is another reason: while the book is a standard fixture in most chemical and physical laboratories, including those in medical centers, it is not as frequently seen in the laboratories of physician's offices (those either in solo or group practice). I believe that the Handbook can be useful in those laboratories. One of the reasons, among others, is that the various basic items of information it offers may be helpful in new tests, either physical or chemical, which are continuously being published. The basic information may relate

Handbook of Chemistry and Physics

Information Theory and Reliable Communication

We present a new technique for the display of high-dynamic-range images, which reduces the contrast while preserving detail. It is based on a two-scale decomposition of the image into a base layer,...

/pdf/fast-bilateral-filtering-for-the-display-of-high-dynamic-43a1vkjsrr.pdf

Fast bilateral filtering for the display of high-dynamic-range images

We have created a system for capturing both the three-dimensional geometry and color and shading information for human facial expressions. We use this data to reconstruct photorealistic, 3D animations of the captured expressions. The system uses a large set of sampling points on the face to accurately track the three dimensional deformations of the face. Simultaneously with the tracking of the geometric data, we capture multiple high resolution, registered video images of the face. These images are used to create a texture map sequence for a three dimensional polygonal face model which can then be rendered on standard 3D graphics hardware. The resulting facial animation is surprisingly life-like and looks very much like the original live performance. Separating the capture of the geometry from the texture images eliminates much of the variance in the image data due to motion, which increases compression ratios. Although the primary emphasis of our work is not compression we have investigated the use of a novel method to compress the geometric data based on principal components analysis. The texture sequence is compressed using an MPEG4 video codec. Animations reconstructed from 512x512 pixel textures look good at data rates as low as 240 Kbits per second.

Making faces

We introduce computational network (CN), a unified framework for describing arbitrary learning machines, such as deep neural networks (DNNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short term memory (LSTM), logistic regression, and maximum entropy model, that can be illustrated as a series of computational steps. A CN is a directed graph in which each leaf node represents an input value or a parameter and each non-leaf node represents a matrix operation upon its children. We describe algorithms to carry out forward computation and gradient calculation in CN and introduce most popular computation node types used in a typical CN. We further introduce the computational network toolkit (CNTK), an implementation of CN that supports both GPU and CPU. We describe the architecture and the key components of the CNTK, the command line options to use CNTK, and the network definition and model editing language, and provide sample setups for acoustic model, language model, and spoken language understanding. We also describe the Argon speech recognition decoder as an example to integrate with CNTK.

/pdf/an-introduction-to-computational-networks-and-the-2kabqb9yg3.pdf

An Introduction to Computational Networks and the Computational Network Toolkit

We exploit the falloff of acuity in the visual periphery to accelerate graphics computation by a factor of 5-6 on a desktop HD display (1920x1080). Our method tracks the user's gaze point and renders three image layers around it at progressively higher angular size but lower sampling rate. The three layers are then magnified to display resolution and smoothly composited. We develop a general and efficient antialiasing algorithm easily retrofitted into existing graphics code to minimize "twinkling" artifacts in the lower-resolution layers. A standard psychophysical model for acuity falloff assumes that minimum detectable angular size increases linearly as a function of eccentricity. Given the slope characterizing this falloff, we automatically compute layer sizes and sampling rates. The result looks like a full-resolution image but reduces the number of pixels shaded by a factor of 10-15.We performed a user study to validate these results. It identifies two levels of foveation quality: a more conservative one in which users reported foveated rendering quality as equivalent to or better than non-foveated when directly shown both, and a more aggressive one in which users were unable to correctly label as increasing or decreasing a short quality progression relative to a high-quality foveated reference. Based on this user study, we obtain a slope value for the model of 1.32-1.65 arc minutes per degree of eccentricity. This allows us to predict two future advantages of foveated rendering: (1) bigger savings with larger, sharper displays than exist currently (e.g. 100 times speedup at a field of view of 70° and resolution matching foveal acuity), and (2) a roughly linear (rather than quadratic or worse) increase in rendering cost with increasing display field of view, for planar displays at a constant sharpness.

Foveated 3D graphics

This paper describes the application of space time constraints to creating transitions between segments of human body motion. The motion transition generation uses a combination of spacetime constraints and inverse kinematic constraints to generate seamless and dynamically plausible transitions between motion segments. We use a fast recursive dynamics formulation which makes it possible to use spacetime constraints on systems with many degrees of freedom, such as human figures. The system uses an interpreter of a motion expression language to allow the user to manipulate motion data, break it into pieces, and reassemble it into new, more complex, motions. We have successfully used the system to create basis motions, cyclic data, and seamless motion transitions on a human body model with 44 degrees of freedom. Additional

/pdf/efficient-generation-of-motion-transitions-using-spacetime-3hbs8mmbzk.pdf

Efficient generation of motion transitions using spacetime constraints

High contrast images are common in night scenes and other scenes that include dark shadows and bright light sources. These scenes are difficult to display because their contrasts greatly exceed the range of most display devices for images. As a result, the image constrasts are compressed or truncated, obscuring subtle textures and details. Humans view and understand high contrast scenes easily, “adapting” their visual response to avoid compression or truncation with no apparent loss of detail. By imitating some of these visual adaptation processes, we developed methods for the improved display of high-contrast images. The first builds a display image from several layers of lighting and surface properties. Only the lighting layers are compressed, drastically reducing contrast while preserving much of the image detail. This method is practical only for synthetic images where the layers can be retained from the rendering process. The second method interactively adjusts the displayed image to preserve local contrasts in a small “foveal” neighborhood. Unlike the first method, this technique is usable on any image and includes a new tone reproduction operator. Both methods use a sigmoid function for contrast compression. This function has no effect when applied to small signals but compresses large signals to fit within an asymptotic limit. We demonstrate the effectiveness of these approaches by comparing processed and unprocessed images.

/pdf/two-methods-for-display-of-high-contrast-images-1fquso4d94.pdf

Brian Guenter

Papers

Making faces

An Introduction to Computational Networks and the Computational Network Toolkit

Foveated 3D graphics

Efficient generation of motion transitions using spacetime constraints

Two methods for display of high contrast images