Abstract We present a simple and effective method for the interactive segmentation of feature regions in a triangular mesh. From the user‐specified radius and click position, the candidate region that contains the desired feature region is defined as geodesic disc on a triangle mesh. A concavity‐aware harmonic field is then computed on the candidate region using the appropriate boundary constraints. An initial isoline is chosen by evaluating the uniformly sampled ones on the harmonic field based on the gradient magnitude. A set of feature points on the initial isoline is selected and the anisotropic geodesics passing through them are then determined as the final segmentation boundary, which is smooth and locally shortest. The experimental results show several segmentation results for various 3D models, revealing the effectiveness of the proposed method.

Current image-to-image translations do not control the output domain beyond the classes used during training, nor do they interpolate between different domains well, leading to implausible results. This limitation largely arises because labels do not consider the semantic distance. To mitigate such problems, we propose a style-aware discriminator that acts as a critic as well as a style encoder to provide conditions. The style-aware discriminator learns a controllable style space using prototype-based self-supervised learning and simultaneously guides the generator. Experiments on multiple datasets verify that the proposed model outperforms current state-of-the-art image-to-image translation methods. In contrast with current methods, the proposed approach supports various applications, including style interpolation, content transplantation, and local image translation. The code is available at github.com/kunheek/style-aware-discriminator.

We propose a method that generates a virtual camera layout of a 3D animation scene by following the cinematic intention of a reference video. From a reference video, cinematic features such as the start frame, end frame, framing, camera movement, and the visual features of the subjects are extracted automatically. The extracted information is used to generate the virtual camera layout, which resembles the camera layout of the reference video. Our method handles stylized as well as human characters with body proportions different from those of humans. We demonstrate the effectiveness of our approach with various reference videos and 3D animation scenes. The user evaluation results show that the generated layouts are comparable to layouts created by the artist, allowing us to assert that our method can provide effective assistance to both novice and professional users when positioning a virtual camera.

Object detection has been applied in a wide variety of real world scenarios, so detection algorithms must provide confidence in the results to ensure that appropriate decisions can be made based on their results. Accordingly, several studies have investigated the probabilistic confidence of bounding box regression. However, such approaches have been restricted to anchor-based detectors, which use box confidence values as additional screening scores during non-maximum suppression (NMS) procedures. In this paper, we propose a more efficient uncertainty-aware dense detector (UADET) that predicts four-directional localization uncertainties via Gaussian modeling. Furthermore, a simple uncertainty attention module (UAM) that exploits box confidence maps is proposed to improve performance through feature refinement. Experiments using the MS COCO benchmark show that our UADET consistently surpasses baseline FCOS, and that our best model, ResNext-64x4d-101-DCN, obtains a single model, single-scale AP of 48.3% on COCO test-dev, thus achieving the state-of-the-art among various object detectors.

We propose Neural Crossbreed, a feed-forward neural network that can learn a semantic change of input images in a latent space to create the morphing effect. Because the network learns a semantic change, a sequence of meaningful intermediate images can be generated without requiring the user to specify explicit correspondences. In addition, the semantic change learning makes it possible to perform the morphing between the images that contain objects with significantly different poses or camera views. Furthermore, just as in conventional morphing techniques, our morphing network can handle shape and appearance transitions separately by disentangling the content and the style transfer for rich usability. We prepare a training dataset for morphing using a pre-trained BigGAN, which generates an intermediate image by interpolating two latent vectors at an intended morphing value. This is the first attempt to address image morphing using a pre-trained generative model in order to learn semantic transformation. The experiments show that Neural Crossbreed produces high quality morphed images, overcoming various limitations associated with conventional approaches. In addition, Neural Crossbreed can be further extended for diverse applications such as multi-image morphing, appearance transfer, and video frame interpolation.

Despite the expected synergistic effects, augmented and virtual reality (AR and VR, respectively) technologies still tend to be discrete entities. In this paper, we describe our effort to enable both AR and VR users to share the same physical geometric space. The geometric transformation between two world spaces, defined separately by AR and VR systems, are estimated using a specially designed tracking board. Once initially obtained, the transformation will allow users to collaborate with each other within an integrated physical environment while making the best use of both AR and VR technologies.

In this paper, we propose an acoustic signal synthesizing algorithm called two-step reconstruction algorithm based on the geometrical information of ROI(region of interest). Even though we apply the same conventional reconstruction algorithm, we can obtain the better image quality of ROI by using the proposed two-step reconstruction algorithm. We comment on the mathematical minor mistakes in applying Residue theorem for the derivation of Green's function in famous paper to which many researchers are still referring. A mathematical k-wave simulation is used for comparing the image quality of ROI with or without two-step reconstruction algorithm. From the simulation results, we prove the effectiveness of proposed acoustic signal synthesizing of two-step reconstruction algorithm.

Adaptive mesh refinement (AMR) is a popular computational simulation technique used in various scientific and engineering fields. Although AMR data is organized in a hierarchical multi-resolution data structure, the traditional volume visualization algorithms such as ray-casting and splatting cannot handle the form without converting it to a sophisticated data structure. In this paper, we present a hierarchical multi-resolution splatting technique using k-d trees and octrees for AMR data that is suitable for implementation on the latest consumer PC graphics hardware. We describe a graphical user interface to set transfer function and viewing/rendering parameters interactively. Experimental results obtained on a general purpose PC equipped with NVIDIA GeForce card are presented to demonstrate that the technique can interactively render AMR data (over 20 frames per second). Our scheme can easily be applied to parallel rendering of time-varying AMR data.

When interactive real-time applications are developed with very large volume data, the use of lossy compression is often inevitable. Lossy compression schemes generally encode data without consideration of the purpose of visualization that is actually performed, which often results in inefficient compression. In this paper, we present a new method for classifying voxels according to their importance in visualization, and assigning appropriate weights to them. The associated weight information can be combined with lossy compression schemes to reduce the visual degradation of reconstructed images, resulting in higher compression rates and visual fidelity. Test results demonstrate that the proposed technique improves both the amount of compression and the quality of visualization significantly.

This paper presents a new scheme for scheduling and control synthesis in high-level circuit design. The scheduling algorithm tries to maximize the performance of a design under resource constraints by maximizing the utilization of resources and minimizing clock slack. It exploits the technique of bit-level chaining (BLC) to target high-speed design. It also exploits noninteger multicycling and chaining, which allows multiple cycle execution of a set of chained operations and even sharing of chained functional units to obtain further performance at the cost of a small increase in the complexity of the control unit. Experimental results on several datapath-intensive designs show significant improvement in throughput over the conventional scheduling algorithms.

This paper presents a new 3D RGB image compression scheme designed for interactive real-time applications. In designing our compression method, we have compromised between two important goals: high compression ratio and fast random access ability, and have tried to minimize the overhead caused during run-time reconstruction. Our compression technique is suitable for applications wherein data are accessed in a somewhat unpredictable fashion, and real-time performance of decompression is necessary. The experimental results on three different kinds of 3D images from medical imaging, image-based rendering, and solid texture mapping suggest that the compression method can be used effectively in developing real-time applications that must handle large volume data, made of color samples taken in three- or higher-dimensional space.

This paper presents a new scheduling algorithm that maximizes the performance of a design under resource constraints in high-level synthesis. The algorithm tries to achieve the maximal utilization of resources and the minimal waste of clock slack time. Moreover, it exploits the technique of bit-level chaining to target high-speed designs. The algorithm tries non-integer multiple-cycling and chaining, which allows multiple cycle execution of chained operations, to further increase the performance at the cost of small increase in the complexity of the control unit. Experimental results on several datapath-intensive designs show significant improvement in execution time, over the conventional scheduling algorithms. 1 Introduction Scheduling in high-level synthesis takes a crucial role of determining the performance of synthesized hardware. As a performance measure, we can use data introduction interval (sample period) or latency when pipelining [1, 2, 3] is not used. Given a data flow g...

While 2D texture mapping is one of the most powerful rendering techniques that make 3D objects appear visually interesting, it suffers from visual artifacts produced when 2D image patterns are wrapped onto the surface of objects with arbitrary shapes. On the other hand, 3D texture mapping generates highly natural visual effects in which objects appear carved from lumps of materials rather than laminated with thin sheets as in 2D texture mapping. Storing 3D texture images in a table for fast mapping computations, instead of evaluating procedures on the fly, however, has been considered impractical due to the extremely high memory requirements. In this paper, we present a new effective method for 3D texture mapping designed for real-time rendering of polygonal models. Our scheme attempts to resolve the potential texture memory problem arising from the very large size of 3D images by compressing them using a wavelet-based encoding method. The experimental results on various non-trivial 3D textures and polygonal models show that high compression rates are achieved with few visual artifacts in the rendered image and a small impact on rendering time. The simplicity of our compression-based scheme will make it possible to implement practical 3D texture mapping in software/hardware rendering systems including the real-time 3D graphics APIs like OpenGL and Direct3D.

Interactive visualization of very large volume data has been recognized as a task requiring great effort in a variety of science and engineering fields. In particular, such data usually places considerable demands on run‐time memory space. In this paper, we present an effective 3D compression scheme for interactive visualization of very large volume data, that exploits the power of wavelet theory. In designing our method, we have compromised between two important factors: high compression ratio and fast run‐time random access ability. Our experimental results on the Visual Human data sets show that our method achieves fairly good compression ratios. In addition, it minimizes the overhead caused during run‐time reconstruction of voxel values. This 3D compression scheme will be useful in developing many interactive visualization systems for huge volume data, especially when they are based on personal computers or workstations with limited memory.