Math and Mechanics in Visual Computing: not too little, not too much （Uri Ascher）

Abstract：Recently there has been a significant move in the visual computing community towards incorporating more mathematically sound techniques and mechanically viable principles into methods and algorithms for computer graphics. Physics-based modelling has often proved to yield significant practical advantages, and automatic data-driven model calibration can be far better than trial and error parameter tuning. Computational techniques from inverse problems and machine learning have been incorporated to yield satisfactory simulations in challenging scenarios. Furthermore, it is increasingly recognized that algorithms without solid justification might look like a collection of hacks and their results might be considered less believable and harder to assess. Mathematical tools may also be used to explain when given algorithms do not work, as well as when they do. Occasionally it is even possible to get insight into physical mechanisms and processes behind visual appearance. This leads to the often difficult passage between qualitative and quantitative results. At the same time, it is also important not to be swayed by sheer mathematical prowess. For one thing, visual computing often requires only physics-based modelling, distinguished from physics modelling and allowing efficient simulation for complex scenarios. Further, insisting on solving differential equations or satisfying mathematical topology theorems can sometime lead to inferior algorithms for visual tasks. Finally, it is advisable to restrain temptation to prove theorems applicable to one algorithm while in fact using another in practical calculations. I will discuss and demonstrate the above statements in the context of specific applications and examples of visual computing, including calibration and simulation of flexible bodies (e.g., cloth), fluid flow (e.g., smoke and water), surface reconstruction and image processing.

Computations and Applications of Medial Axis Transform of 3D Shapes （Wenping Wang）

Abstract：As a complete shape description, the medial axis of a geometric shape possesses a number of favorable properties--it encodes symmetry, local thickness and structural components of the shape it represents. Hence, the medial axis has been studied extensively in shape modeling and analysis since its introduction by Blum in 1960s. However, the practical application of the medial axis is hindered by its notorious instability and lack of compact representation; that is, a primitive medial axis without proper processing is often represented as a dense discrete mesh with many spurious branches. In this talk I shall represent some recent studies on computing stable and compact representations of the medial axes of 3D shapes. Techniques from mesh simplification will be extended to compute a medial axis without spurious branches and represented by a small number of mesh vertices, while meeting specified approximation accuracy. New applications of the medial axis will also be presented.

Mapping the World with Drones （Long Quan）　

Abstract：In the first part of the talk, I will review the state of the art of the three dimensional reconstruction from images or photographs developed in the past three decades in computer vision. In the second part of the talk, I will focus on the most recent exciting work of large-scale 3D reconstruction from drone photographs, and showcase the performances of our approach over a large samples of case studies of hundreds square kilometres in both high-rise metropolitan areas and low-rise rural areas in different cities of different countries. I will also demonstrate the online cloud platform and portal www.altizure.com, developed and funded by the HKUST team.

Closing the Loop: from acquisition to manufacturing 3D （Andrei Sharf）　

Abstract：The evolution of 3D scanning technologies have revolutionized the way real-world object are digitally acquired. Nowadays, highdefinition and high-speed scanners can capture even large scale scenes with very high accuracy. Nevertheless, the acquisition of complete 3D objects remains a bottleneck, requiring to carefully sample the whole object’s surface, similar to a coverage process. In parallel we are witnessing in recent years a growing interest in 3D printing technologies, enabling the physical realization of a large variety of shape classes. In general, printability of a 3D model requires the model to be structurally sound, with geometric details that are larger than the printer’s resolution and to be self-supported.This is especially challenging in the case of shapes that involve complex structures and fine details.In this talk, I will attempt to take a tour from acquisition to 3D manufacturing, showing some of the challenges and solutions that lie amidst.

Data Driven Approach in Computer Graphics （Xin Tong）

Abstract：traditionally, the graphics algorithms follow the physical rules to simulate the dynamics of objects and human being and compute the light transport in a scene. Although this method could provide high realism results, the computational cost is always high. In this talk, I will introduce how to apply a new data driven approach to tackle different graphics problems. By combining the carefully captured data and machine learning techniques, the data driven approach efficiently reduce the computational cost and preserve the result quality.

Gradient Domain Manipulation （Dani Lischinski）　

Abstract：Gradient domain manipulation is an important technique with multiple applications in image and video tone mapping and editing. In this talk, I will explain the relevant underlying mathematical theory and show how operating in the gradient domain can be used for tone mapping of High Dynamic Range (HDR) images, as well as for seamless image cloning and stitching.

Computer Vision that Mimics and Surpasses Human Ability （Jiaya Jia）　

Abstract：This talk covers general review of computer vison research in recent years from two perspectives. It will be first exemplified by the computer vision goals that cannot be easily achieved by human. These tasks involve solving a series of low-level problems such as filtering, stereo matching, depth estimation, deconvolution, and motion estimation. Then a few hot topics to simulate human intelligence in image understanding will be introduced, which include semantic segmentation, object classification, and object detection. Several techniques developed in our team will be demonstrated.

Cross-modal shape retrieval using semantic-based descriptors （Flora Tasse）　

Abstract：Convolutional neural networks (CNNs) have been successfully used to compute shape descriptors, or jointly embed shapes and sketches in a common vector space. An alternative approach to this is to construct a fixed intermediate space such as a vector space of words and then, map an arbitrary number of different modalities into that space. I will first describe a method that uses natural language processing techniques and state-of-art in multi-view CNNs to compute semantic-based descriptors for shapes, sketches, natural images and range scans. I will then show a few applications such as cross-modal shape retrieval and zero-shot learning.

Pattern-Aware Selection and Arrangement of Graphic Elements （Pengfei Xu ）

Abstract： Selection and arrangement of graphic elements are two fundamental tasks in many text and graphic editing scenarios. Traditional interactive techniques such as lasso selection, snapping and arrangement commands are provided by most graphic editors to aid the users in accomplishing these tasks. However, these techniques often ignore the underlying patterns of the elements, requiring the users to perform explicit and tedious operations to achieve the goal of the manipulation. It is expected that the amount of user interaction can be reduced by exploiting the patterns of the elements. Nevertheless, designing effective pattern-aware interaction tools is challenging due to the ambiguities in the patterns presented by the elements and the operations issued by the user. In this talk, I will introduce three effective pattern-aware interaction techniques to aid the users in the selection and arrangement of elements. In particular, I will introduce 1) Lazy Selection, a scribble-based tool for quick selection of one or more desired shape elements by roughly stroking through the elements, 2) GACA, a group-aware command-based arrangement tool for arranging multiple groups of elements with a single command, and 3) A framework for automatic global beautification of layouts of graphic elements with gestural interface for editing the patterns of the elements. These three techniques adopt different strategies to resolve the ambiguities that may arise during manipulation, ensuring their usability.

Interaction-Based Functionality Analysis of 3D Shapes （Ruizhen Hu ）

Abstract：The majority of man-made objects are designed to serve a certain function, and this is often reflected by the geometry of the objects, or the way that they are used and organized in an environment. In recent years, many efforts in shape analysis have developed methods that extract high-level structural and semantic information from geometric shapes and scenes, especially involving man-made objects. One can argue that the ultimate goal of some of these works is to understand the functionality of the objects, since functionality considerations can aid in applications such as semantic classification, editing, and synthesis. Moreover, there have also been works that more explicitly model and incorporate functionality into the processing of shapes and scenes. Thus, functionality has been receiving increasingly more attention in shape analysis and geometric modeling. In this talk, I will talk about recent developments that incorporate functionality aspects in the analysis of 3D shape, especially those based on interactions.

Intrinsic Decomposition, Salient Object Detection and Discrete Geodesic Computation （Xiaoguang Han）

Abstract：The talk includes three parts: edge-aware image smoothing and intrinsic decomposition, image salient object detection and discrete geodesic computation. Identifying sparse salient structures from dense pixels is a long-standing problem in visual computing. In the first part, I introduce a piece-wise image flattening method for edge-preserving smoothing and scene-level intrinsic decomposition which delivers the best results to date on IIW database (see our paper “An L1 Image Transform for Edge-Preserving Smoothing and Scene-Level Intrinsic Decomposition” in SIGGRAPH 2015). Visual saliency serves as an important pre-processing step for a variety of content-aware image editing tasks, including image cropping, retargeting, and summarization. In the second part, I show you a deep learning framework for salient object detection which includes the best-performing algorithms to date on existing benchmark datasets (see our CVPR 2015&2016 papers, “Visual Saliency Based on Multiscale Deep Features” and “Deep Contrast Learning for Salient Object Detection”). Computing discrete geodesic distance over triangle meshes is one of the fundamental problems in computational geometry and computer graphics. In the last part of this talk, I introduce a new geodesic computation framework based on triangle-oriented window propagation which is the fastest method to find exact geodesic distance and path on triangular meshes (see our paper “Fast and Exact Discrete Geodesic Computation Based on Triangle-Oriented Wavefront Propagation” in SIGGRAPH 2016).

Data Visualization: an introduction （Oliver Deussen）　

Abstract：In data visualization large amounts of data are converted into images in order to use the human eye to find interesting patterns that can later be analysed by data mining techniques. In my talk I will give brief overview about various data visualization techniques and will mention the most important research questions of this field.

Computational Interlocking Methods for Fabricating 3D Assembly （Chi-Wing FU）　

Abstract：Constructing a 3D object assembly is an important topic of great interest in computer graphics research, particularly in the areas of 3D printing, mechanical design, and furniture modeling. The research problem involves the construction of 3D component parts, the construction of joints for connecting the component parts, as well as many other issues such as stability and aesthetics. To create the component connections in making 3D assembly, we first propose to employ the concept of mechanical interlocking found in the well-known puzzle model known as the Burr puzzle; the interesting property of this model is that in a finished assembly, all the component parts are immobilized by the geometry, except for a special key. Hence, we can tighten the connections among the component parts in the assembly. Starting from this puzzle model, we generalize the interlocking concept, and formulate various novel computational methods for constructing mechanical interlocking in different 3D fabrication forms, including our preliminary work in the recursive interlocking puzzle models, later the local-interlocking group concept for computing the joints in furniture parts, and more recently the corner-based interlocking mechanism for interlocking 2D laser-cut parts into a 3D polyhedral object.

Robot Explorative 3D Sensing and Understanding （Kai Xu ）

Abstract：Visual sensing is perhaps the most important channel for a robot to understand the world. With the fast development of 3D sensing techniques, and the proliferation of 3D geometric databases, computer graphics is being deeply fused with computer vision, leading to 3D-geometry-based vision which provides substantial support to modern robot vision. We study robot-operated explorative 3D geometry sensing and understanding, which admits 3D input and is driven by 3D geometric data. We first discuss some general design considerations for such algorithms and systems. We then introduce two of our recent works, including robot scene scanning with pro-active object analysis, and 3D attention-based fine-grained object recognition.

3D Printing Oriented Geometric Design and Optimization （Ligang Liu）

Abstract：3D printers have become popular in recent years and enable fabrication of custom objects for home users. The promise of moving creations from a virtual space into reality is truly tantalizing, and its applications go far beyond basic manufacturing and rapid prototyping. However, many obstacles remain for 3D printing to be practical and commonplace. In this talk, I will review our recent works on geometric modeling and processing for 3D printing applications.

Geometric Decomposition for 3D Shapes （Yang Zhou）

Abstract：Decomposing a complex shape into geometrically simple primitives is a fundamental problem in geometry processing. In this talk, I’ll introduce you 3 different geometric decomposition methods for 3D shapes, where the simple primitives are approximate convex, pyramid and generalized cylinder respectively. Though aiming at different goals, all these 3 methods are based on bottom-up strategies, while the last two further optimize the decomposition with a global objective. Rather than introducing the details of how they are implemented, I’ll focus on presenting what these works about and why we need them, i.e. what these methods can be used for.

A Deep Learning Framework for Character Motion Synthesis and Editing （Taku Komura ）

Abstract：We present a framework to synthesize character movements based on high level parameters, such that the produced movements respect the manifold of human motion, trained on a large motion capture dataset. The learned motion manifold, which is represented by the hidden units of a convolutional autoencoder, represents motion data in sparse components which can be combined to produce a wide range of complex movements. To map from high level parameters to the motion manifold, we stack a deep feedforward neural network on top of the trained autoencoder. This network is trained to produce realistic motion sequences from parameters such as a curve over the terrain that the character should follow, or a target location for punching and kicking. The feedforward control network and the motion manifold are trained independently, allowing the user to easily switch between feedforward networks according to the desired interface, without re-training the motion manifold. Once motion is generated it can be edited by performing optimization in the space of the motion manifold. This allows for imposing kinematic constraints, or transforming the style of the motion, while ensuring the edited motion remains natural. As a result, the system can produce smooth, high quality motion sequences without any manual pre-processing of the training data.

Texture Synthesis and Weathering （Daniel Cohen-Or）　

Abstract：I will present a new technique to synthesize time-varying weathered textures. Given a single image as input, the degree of weathering at different regions of the input image is estimated by prevalence analysis of the texture patches. This information allows then to gracefully increase or decrease the popularity of weathered patches. This simulates the evolution of texture appearance both backward and forward in time. This method can be applied to a wide variety of different textures since the material’s reaction to weathering effects are physically-oblivious and learned from the input texture itself. In my talk, I will start with a rather long introduction to texture synthesis. I will present the basic ideas of non-parametric texture synthesis, quilting, texture transfer, and image completion.

Challenges forVisualization in Data Technology (DT) Era （Wei Chen）

Abstract：Despite that data visualization has been applied for many fields, there remains many challenges. First, data analysis typically search known patterns from data. How to predict unexpected information remains a challenging problem. Second, despite many well-established software, toolkits and systems, a comprehensive visualization scheme and strategy that is adaptable to varied tasks and datasets is demanded. However, a general visualization software or system is impractical. Thus, we need to first standardize a visualization and visualization design. The standardization requires further revolutions for effectively improving the effectiveness and efficiency of visualization. Third, urban data is connected to cyber-and-physical spaces. To address practical problems in urban management and security, a suite of schemes are demanded for realizing urban data visualization.

Editable and Compact Representation of 3D Animated Meshes （Guiqing Li ）

Abstract：Compactly representing time-varying geometries is an important issue in dynamic geometry processing. This talk studies a kind of blend shape representations which are derived through analyzing the shape space defined by a set of 3D animation sequences. We will first briefly review a few typical representations and their applications then introduce a sparse localized decomposition of animated meshes by using edge lengths and dihedral angles to capture shape and pose variations of animated meshes.

Co-segmentation of Sets of Shapes and Applications （Oliver van Kaick）　

Abstract：During the last decade, research in computer graphics has given much attention to the problem of facilitating shape modeling and the creation of 3D content. Modeling requires artists to be highly skilled and typically involves considerable work. Thus, a recent research trend is to develop tools that facilitate modeling by manipulating 3D shapes at a higher-level, especially by representing shapes as collections of parts (such as the legs, seat and back of a chair). To create such high-level representations, we first need to analyze the shapes and discover their part decomposition. In this context, a series of works that have been successful in deriving part representations for shapes are based on co-analysis. The main idea of co-analysis is that more knowledge can be inferred by analyzing a set of shapes in conjunction, rather than analyzing each shape individually. By extending this principle to segmentation, we arrive at co-segmentation, where we use the knowledge latent in a set of shapes to obtain a more semantic, consistent segmentation of the shapes. In this talk, I will first present the progress in co-segmentation by discussing representative approaches. Then, I will present a few content creation applications that benefit from a consistent segmentation of a set, such as modeling with the aid of a meta-representation.

Surface Mosaic Synthesis with Irregular Tiles （Zhonggui Chen ）

Abstract：Mosaics are widely used for surface decoration to produce appealing visual effects. We present a method for synthesizing digital surface mosaics with irregularly shaped tiles, which are a type of tiles often used for mosaics design. Our method employs both continuous optimization and combinatorial optimization to improve tile arrangement. In the continuous optimization step, we iteratively partition the base surface into approximate Voronoi regions of the tiles and optimize the positions and orientations of the tiles to achieve a tight fit. Combination optimization performs tile permutation and replacement to further increase surface coverage and diversify tile selection. The alternative applications of these two optimization steps lead to rich combination of tiles and high surface coverage.