PG2019 Short Papers

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/2632814

Browse

Now showing 1 - 10 of 10

LPaintB: Learning to Paint from Self-Supervision
(The Eurographics Association, 2019) Jia, Biao; Brandt, Jonathan; Mech, Radomír; Kim, Byungmoon; Manocha, Dinesh; Lee, Jehee and Theobalt, Christian and Wetzstein, Gordon
We present a novel reinforcement learning-based natural media painting algorithm. Our goal is to reproduce a reference image using brush strokes and we encode the objective through observations. Our formulation takes into account that the distribution of the reward in the action space is sparse and training a reinforcement learning algorithm from scratch can be difficult. We present an approach that combines self-supervised learning and reinforcement learning to effectively transfer negative samples into positive ones and change the reward distribution. We demonstrate the benefits of our painting agent to reproduce reference images with brush strokes. The training phase takes about one hour and the runtime algorithm takes about 30 seconds on a GTX1080 GPU reproducing a 1000x800 image with 20,000 strokes.
3D Human Body Skeleton Extraction from Consecutive Surfaces
(The Eurographics Association, 2019) Zhang, Yong; Tan, Fei; Wang, Shaofan; Kong, Dehui; Yin, Baocai; Lee, Jehee and Theobalt, Christian and Wetzstein, Gordon
Extracting human body skeletons from consecutive surfaces is an important research topic in the fields of computer graphics and human computer interaction, especially in posture estimation and skeleton animation. Current approaches mainly suffer from following problems: insufficient time and space continuity, not robust to background, ambient noise, etc. Our approach is to improve against these shortcomings. This paper proposes a 3D human body skeleton extraction method from consecutive meshes. We extract the consistent skeletons from consecutive surfaces based on shape segmentation and construct skeleton sequences, then we use the continuous frame skeleton point optimization model we proposed to optimize the skeleton sequences, generating the final skeleton point sequences which are more accurate. Finally, we verify that our method can obtain more complete and accurate skeletons compared to other methods through many experiments.
Gaze Attention and Flow Visualization using the Smudge Effect
(The Eurographics Association, 2019) Yoo, Sangbong; Jeong, Seongmin; Kim, Seokyeon; Jang, Yun; Lee, Jehee and Theobalt, Christian and Wetzstein, Gordon
Many advanced gaze visualization techniques have been developed continuously based on the fundamental gaze visualizations such as scatter plots, attention map, and scanpath. However, it is not easy to locate challenging visualization techniques that resolve the limitations presented in the conventional gaze visualizations. Therefore, in this paper, we propose a novel visualization applying the smudge technique to the attention map. The proposed visualization intuitively shows the gaze flow and AoIs (Area of Interests) of an observer. Besides, it provides fixation, saccade, and micro-movement information, which allows us to respond to various analytical goals within a single visualization. Finally, we provide two case studies to show the effectiveness of our technique.
Towards Biomechanically and Visually Plausible Volumetric Cutting Simulation of Deformable Bodies
(The Eurographics Association, 2019) Qian, Yinling; Huang, Wenbin; Si, Weixin; Liao, Xiangyun; Wang, Qiong; Heng, Pheng-Ann; Lee, Jehee and Theobalt, Christian and Wetzstein, Gordon
Due to the simplicity and high efficiency, composited finite element method(CFEM) based virtual cutting attracted much attention in the field of virtual surgery in recent years. Even great progress has been made in volumetric cutting of deformable bodies, there are still several open problems restricting its applications in practical surgical simulator. First among them is cutting fracture modelling. Recent methods would produce cutting surface immediately after an intersection between the cutting plane and the object. But in real cutting, biological tissue would first deform under the external force induced by scalpel and then fracture occurs when the stress exceeds a threshold. Secondly, it's computation-intensive to reconstruct cutting surface highly consistent with the scalpel trajectory, since reconstructed cutting surface in CFEM-based virtual cutting simulation is grid-dependent and the accuracy of cutting surface is proportional to the grid resolution. This paper propose a virtual cutting method based on CFEM which can effectively simulate cutting fracture in a biomechanically and visually plausible way and generate cutting surface which is consistent with the scalpel trajectory with a low resolution finite element grid. We model this realistic cutting as a deformation-fracture repeating process. In deformation stage, the object will deform along with the scalpel motion, while in the fracture stage cutting happens and a cutting surface will be generated from the scalpel trajectory. A delayed fracturing criteria is proposed to determine when and how the cutting fracture occurs and an influence domain adaptation method is employed to generate accurate cutting surface in both procedures of deformation and fracture. Experiments show that our method can realistically simulate volumetric cutting of deformable bodies and efficiently generate accurate cutting surface thus facilitating interactive applications.
A Psychophysical Analysis of Fabricated Anisotropic Appearance
(The Eurographics Association, 2019) Filip, Jiri; Kolafová, Martina; Vávra, Radomir; Lee, Jehee and Theobalt, Christian and Wetzstein, Gordon
Many materials change surface appearance when observed for fixed viewing and lighting directions while rotating around its normal. Such distinct anisotropic behavior manifests itself as changes in textural color and intensity. These effects are due to structural elements introducing azimuthally-dependent behavior. However, each material and finishing technique has its unique anisotropic properties which are often difficult to control. To avoid this problem, we study controlled anisotropic appearance introduced by means of 3D printing. Our work tends to link perception of directionality with perception of anisotropic reflectance effect it causes. We simulate two types of structure-based anisotropic effects, which are related to directional principles found in real-world materials. For each type, we create a set of test surfaces by controlling the printed anisotropy level and assess them in a psychophysical study to identify a perceptual scale of anisotropy. The generality of these scales is then verified by means of anisotropic surfaces appearance capturing using bidirectional texture function and its analysis on 3D objects. Eventually, we relate the perceptual scale of anisotropy to a computational feature obtained directly from anisotropic highlights observed in the captured reflectance data. The feature is validated using a psychophysical study analyzing visibility of anisotropic reflectance effects.
Connectivity-preserving Smooth Surface Filling with Sharp Features
(The Eurographics Association, 2019) Lescoat, Thibault; Memari, Pooran; Thiery, Jean-Marc; Ovsjanikov, Maks; Boubekeur, Tamy; Lee, Jehee and Theobalt, Christian and Wetzstein, Gordon
We present a method for constructing a surface mesh filling gaps between the boundaries of multiple disconnected input components. Unlike previous works, our method pays special attention to preserving both the connectivity and large-scale geometric features of input parts, while maintaining efficiency and scalability w.r.t. mesh complexity. Starting from an implicit surface reconstruction matching the parts' boundaries, we first introduce a modified dual contouring algorithm which stitches a meshed contour to the input components while preserving their connectivity. We then show how to deform the reconstructed mesh to respect the boundary geometry and preserve sharp feature lines, smoothly blending them when necessary. As a result, our reconstructed surface is smooth and propagates the feature lines of the input. We demonstrate on a wide variety of input shapes that our method is scalable to large input complexity and results in superior mesh quality compared to existing techniques.
RegionSketch: Interactive and Rapid Creation of 3D Models with Rich Details
(The Eurographics Association, 2019) Liu, Shuai; Hou, Fei; Hao, Aimin; Qin, Hong; Lee, Jehee and Theobalt, Christian and Wetzstein, Gordon
In this paper, we articulate a new approach to interactive generation of 3D models with rich details by way of sketching sparse 2D strokes. Our novel method is a natural extension of Poisson vector graphics (PVG). We design new algorithms that distinguish themselves from other existing sketch-based design systems with three unique features: (1) A novel sketch metaphor to create freeform surface based on Poisson's equation, which is simple, intuitive, and free of ambiguity; (2) Convenient and flexible user interface that affords the user to add rich details to the surface with simple sketch input; and (3) Rapid model creation with sparse strokes, which enables novice users to enjoy the utilities of our system to create expected 3D models. We validate the proposed method through a large repository of interactively sketched examples. Our experiments and produced results confirm that our new method is a simple yet efficient design tool for modeling free-form shapes with simple and intuitive 2D sketches input.
Feature Curve Network Extraction via Quadric Surface Fitting
(The Eurographics Association, 2019) Zhengda, Lu; Guo, Jianwei; Xiao, Jun; Wang, Ying; Zhang, Xiaopeng; Yan, Dong-Ming; Lee, Jehee and Theobalt, Christian and Wetzstein, Gordon
Feature curves on 3D shapes provide a high dimensional representation of the geometry and reveal their underlying structure. In this paper, we present an automatic approach for extracting complete feature curve networks from 3D models, as well as generating a high-quality patch layout. Starting from an initial collection of noisy and fragmented feature curves, we first filter non-salient or noisy feature curves by utilizing a quadric surface fitting technique. We then handle the curve intersections and curve missing by conducting a feature extension step to form a closed feature curve network. Finally, we generate a patch layout to reveal a highly structured representation of the input surfaces. Experimental results demonstrate that our algorithm is robust for extracting complete feature curve networks from complex input meshes and achieves superior quality patch layouts compared with the state-of-the-art approaches.
External Forces Guided Fluid Surface and Volume Reconstruction from Monocular Video
(The Eurographics Association, 2019) Nie, Xiaoying; Hu, Yong; Su, Zhiyuan; Shen, Xukun; Lee, Jehee and Theobalt, Christian and Wetzstein, Gordon
We propose a novel method to reconstruct fluid's volume movement and surface details from just a monocular video for the first time. Although many monocular video-based reconstruction methods have been developed, the reconstructed results are merely one layer of geometry surface and lack physically correct volume particles' attribute and movement. To reconstruct 3D fluid volume, we define two kinds of particles, the target particles and the fluid particles. The target particles are extracted from the height field of water surface which is recovered by Shape from Shading (SFS) method. The fluid particles represent the discrete form of the 3D fluid volume and conform to the flow hydrodynamic properties. The target particles are used to guide the physical simulation of fluid particles based on the Smoothed Particle Hydrodynamics (SPH) model. To formulate this guidance, a new external force scheme is designed based on distance and relative motion between target particles and fluid particles. Additionally, in order to integrate and maintain geometric and physical features simultaneously, we adopt a two-scale decomposition strategy for the height field, and only apply the low frequency coarse-scale component to estimate the volumetric motion of liquid, while serve high frequency fine-scale component as noise to preserve fluid surface details in the stage of rendering. Our experimental results compare favorably to the state-of-the-art in terms of global fluid volume motion features and fluid surface details and demonstrate our approach can achieve desirable and pleasing effects.
Pacific Conference on Computer Graphics and Applications - Short Papers 2019: Frontmatter
(Eurographics Association, 2019) Lee, Jehee; Theobalt, Christian; Wetzstein, Gordon; Lee, Jehee and Theobalt, Christian and Wetzstein, Gordon

Browse

Browsing PG2019 Short Papers by Issue Date

Results Per Page

Sort Options