Convolutional neural networks (CNNs) have achieved success in optical flow estimation using labeled datasets, but they fail to build an internal representation to fast adapt to the specific task. On the other hand, limited by the lack of ground truth, existing CNNs-based methods suffer from high noise sensitivity and inferior generalization performance. We integrate the meta-learning technique with optical flow estimation, which can learn internal features to search optimal initial state parameters of the network. Meanwhile, we devise an enhanced network termed MetaFlow to further improve performance. MetaFlow extracts per-pixel features, builds correlation volumes for all pairs of pixels, and iteratively updates optical flow through optical flow predictor using meta-learning. In addition, we propose a meta-transfer pretraining approach to obtain initial network weights, which can efficiently avoid network overfitting. Empirical experiments on MPI Sintel and KITTI benchmarks have shown that the proposed MetaFlow achieves the state-of-the-art results and performs outstanding in challenging scenarios such as textureless regions and abrupt motions.
KEYWORDS: Visual process modeling, Virtual reality, Data modeling, Solid modeling, Motion models, Human vision and color perception, Databases, Data processing, Vision-based navigation, Computer simulations
Intelligent virtual human is widely required in computer games, ergonomics software, virtual environment and so on. We present a vision-based behavior modeling method to realize smart navigation in a dynamic environment. This behavior model can be divided into three modules: vision, global planning and local planning. Vision is the only channel for smart virtual actor to get information from the outside world. Then, the global and local planning module use A* and
D* algorithm to find a way for virtual human in a dynamic environment. Finally, the experiments on our test platform (Smart Human System) verify the feasibility of this behavior model.
Object-order volume rendering algorithms play important part in many visualization applications for their excellent performances. Though many volume rendering algorithms have been proposed during the past two decades, most of them are image-order algorithms. Splatting, one of the classical object-order algorithms, suffers from several kinds of aliasing artifacts for inaccuracy reasons. A much accurate object-order volume rendering algorithm is presented in this paper. By defining a set of data structures to serve as two step reconstruction lookup tables, together with using a simple voxel traversal and resample strategy, the new algorithm can not only get rid of inaccuracy of traditional splatting, but also have the features including high cache hit rate, easy to implement of parallelism and high speedup from pre-processing.
This paper described the design of the DAVRS system. This system not only provides a 3D design environment for architects, but also realizes distributed collaboration between the designers through the internet. The DAVRS system used Java3D to construct virtual sense, XML to pocket sense controlling data, and Java MQ to transmit data. Moreover, a three-level distributed collaboration model is designed in order to confirm the safety of collaboration based on internet.
Volume rendering has been a key technology in the visualization of data sets from various disciplines. However, real-time volume rendering of large scale data sets is still a challenging field due to the vast memory, bandwidth and computational requirements. In this paper, to visualize small to medium scale data set in real-time, we first proposed a new kind of volume rendering graphic processor based on object-order splatting algorithm in which flexible transfer function configuration and software optimization such as early opacity termination and transparent voxel occlusion can be achieved. At the same time, the processor also integrates an eight-way interleaved memory system and an efficient address calculation module to accelerate the voxel traversal process and maintain high cache hit rate. Multiple parallel rendering pipelines embedded also can achieve local parallelism on board. Second, in order to render large scale data sets, a real-time and general-purpose volume rendering architecture is also presented in this paper. By utilizing graphic processors on PC clusters, large scale data sets can be visualized resulted from the high parallel speedup among graphic processors.
An architecture of distributed virtual reality system is brought forward in this paper. It's based on a C/S system model, and employs a centralized-and-distributed data distribution model. This data distribution model can efficiently realize concurrency control, and easily ensure the data consistency. It also employs a four-layered structure based on Message Queue to accomplish the collaboration management. This structure is platform-independent and very flexible for further extending and upgrading. In this paper, we'll introduce the C/S system model and the centralized-and-distributed data distribution model, and then discuss in detail how the four-layered structure based on Message Queue realizes the collaboration management.
Virtual human is widely used in educational, entertainment, and 3D game software. In this paper, the subsumption architecture, which is popular in robotics, is employed to make a two-level control model. With this model, virtual human can realize collision-free navigation in a dynamic environment. As collision avoidance is realized in the 0-level layer, path planning is completed in the 1-level layer. Finally, an experiment was done in our human animation platform. And the result shows that this control model can guide digital actor navigate collision-freely in the environment with dynamic and static obstacles.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.