Recent deep networks that directly handle points in a point set, e.g., PointNet, have been state-of-the-art for supervised learning tasks on clouds such as classification and segmentation. In this work, novel end-to-end auto-encoder is proposed to address unsupervised challenges clouds. On the encoder side, graph-based enhancement enforced promote local structures top of PointNet. Then, folding-based decoder deforms canonical 2D grid onto underlying 3D object surface cloud, achieving low...

Unlike on images, semantic learning 3D point clouds using a deep network is challenging due to the naturally unordered data structure. Among existing works, PointNet has achieved promising results by directly sets. However, it does not take full advantage of point's local neighborhood that contains fine-grained structural information which turns out be helpful towards better learning. In this regard, we present two new operations improve with more efficient exploitation structures. The first...

It is challenging to measure the geometry distortion of point cloud introduced by compression. Conventionally, errors between clouds are measured in terms point-to-point or point-to-surface distances, that either ignores surface structures heavily tends rely on specific reconstructions. To overcome these drawbacks, we propose using point-to-plane distances as a geometric distortions The intrinsic resolution proposed normalizer convert mean square PSNR numbers. In addition, perceived local...

To facilitate new video applications such as three-dimensional (3DV) and free-viewpoint (FVV), multiple view plus depth format (MVD), which consists of both views the corresponding per-pixel images, is being investigated. Virtual can be generated using image based rendering (DIBR), takes images input. This paper discusses synthesis techniques on DIBR, includes forward warping, blending hole filling. Especially, we will emphasize brought to MPEG reference software (VSRS). Unlike case in field...

To reduce cost in storing, processing and visualizing a large-scale point cloud, we consider randomized resampling strategy to select representative subset of points while preserving application-dependent features.The proposed is based on graphs, which can represent underlying surfaces lend themselves well efficient computation.We use general feature-extraction operator features propose reconstruction error evaluate the quality resampling.We obtain form optimal distribution by minimizing...

Video representations that support view synthesis based on depth maps, such as multiview plus (MVD), have been recently proposed, raising interest in efficient tools for map coding. In this paper, we derive a new distortion metric takes into consideration camera parameters and global video characteristics order to quantify the effect of lossy coding maps synthesized quality. addition, skip mode selection method is proposed local characteristics. Experimental results with scheme show gains up...

Geometric data acquired from real-world scenes, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e.g.</i> , 2D depth images, 3D point clouds, and 4D dynamic have found a wide range of applications including immersive telepresence, autonomous driving, surveillance, xmlns:xlink="http://www.w3.org/1999/xlink">etc</i> . Due to irregular sampling patterns most geometric data, traditional image/video processing methodologies are limited, while Graph...

New data formats that include both video and the corresponding depth maps, such as multiview plus (MVD), enable new applications in which intermediate views (virtual views) can be generated using transmitted/stored (reference maps inputs. We propose a map coding method based on distortion measurement by deriving relationships between distortions coded rendered view. In our experiments we use codec H.264/AVC tools, where rate-distortion (RD) optimization for encoding makes of metric. Our...

With the development of 3D display and interactive multimedia systems, new video applications, such as 3DTV Free Viewpoint Video, are attracting significant interests. In order to enable these data formats including captured 2D sequences corresponding depth maps have been proposed. Compared conventional frames, very different characteristics. First, they typically consist homogeneous areas partitioned by sharp edges representing discontinuities, while discontinuities play important roles in...

With the recent improvements in 3-D capture technologies for applications such as virtual reality, preserving cultural artifacts, and mobile mapping systems, new methods compressing point cloud representations are needed to reduce amount of bandwidth or storage consumed. For clouds having attributes color associated with each point, several existing perform attribute compression by partitioning into blocks reducing redundancies among adjacent points. If, however, many sparsely populated, few...

Point clouds are becoming essential in key applications with advances capture technologies leading to large volumes of data. Compression is thus for storage and transmission. In this work, the state art geometry attribute compression methods a focus on deep learning based approaches reviewed. The challenges faced when compressing attributes considered, an analysis current address them, their limitations relations between traditional ones. Current open questions point cloud compression,...

High delivery ratio with low energy consumption is one of design challenges for wireless sensor network routing protocols. In this paper, we identify the drawbacks pure single path scheme and multipath scheme, in terms guaranteed consumption. Accordingly, describe a which data forwarded along pre-established to save energy, high achieved by repair whenever break detected. We propose simple, quick, local repairing approach, whereby pivot node can skip over only using already existing...

3D Video (3DV) with depth-image-based view synthesis is a promising candidate of next generation broadcasting applications. However, the synthesized views in 3DV are often contaminated by annoying artifacts, particularly notably around object boundaries, due to imperfect depth maps (e.g., produced state-of-the-art stereo matching algorithms or compressed lossily). In this paper, we first review some representative methods for boundary artifact reduction synthesis, and make an in-depth...

With the increased proliferation of applications using 3-D capture technologies for such as virtual reality, mobile mapping, scanning historical artifacts, and printing, representing these kinds data 3-Dpoint clouds has become a popular method storing conveying independently how it was captured. A point cloud consists set coordinates indicating location each point, along with one or more attributes color associated point. Because size can be quite large, compression is needed to efficiently...

Scene flow depicts the dynamics of a 3D scene, which is critical for various applications such as autonomous driving, robot navigation, AR/VR, etc. Conventionally, scene estimated from dense/regular RGB video frames. With development depth-sensing technologies, precise measurements are available via point clouds have sparked new research in flow. Nevertheless, it remains challenging to extract due sparsity and irregularity typical cloud sampling patterns. One major issue related irregular...

Point cloud compression (PCC) is a key enabler for various 3-D applications, owing to the universality of point format. Ideally, 3D clouds endeavor depict object/scene surfaces that are continuous. Practically, as set discrete samples, locally disconnected and sparsely distributed. This sparse nature hindering discovery local correlation among points compression. Motivated by an analysis with fractal dimension, we propose heterogeneous approach deep learning lossy geometry On top base layer...

We propose an analytical model to estimate the synthesized view quality in 3D video. The relates errors depth images synthesis quality, taking into account texture image characteristics, and rendering process. Especially, we decompose distortion texture-error induced depth-error distortion. analyze using approach combining frequency spatial domain techniques. Experiment results with video sequences coding/rendering tools used MPEG 3DV activities show that our can accurately noise power....

Depth images are often presented at a lower spatial resolution, either due to limitations in the acquisition of depth or increase compression efficiency. As result, upsampling low-resolution higher resolution is typically required prior image based rendering. In this paper, enhancement and up-sampling techniques proposed using graph-based formulation. one scheme, first upsampled conventional method, then followed by joint bilateral filtering enhance edges reduce noise. A second scheme avoids...

In order to improve 3D video coding efficiency, we propose methods estimate rendered view distortion in synthesized views as a function of the depth map quantization error. Our approach starts by calculating geometric error caused based on camera parameters. Then, local characteristics. The estimated is used rate-distortion optimized mode selection for coding. A Lagrange multiplier derived using proposed metric, which an autoregressive model. Experimental results show efficiency methods,...

Topology matters. Despite the recent success of point cloud processing with geometric deep learning, it remains arduous to capture complex topologies data a learning model. Given dataset containing objects various genera, or scenes multiple objects, we propose an autoencoder, TearingNet, which tackles challenging task representing clouds using fixed-length descriptor. Unlike existing works directly deforming predefined primitives genus zero (e.g., 2D square patch) object-level cloud, our...

New data formats including 2D video and the corresponding depth maps enable new applications in which virtual views can be rendered, such as 3DTV free-viewpoint (FVV). Different from frames, typically consist of homogeneous areas (with no textures) separated by sharp edges representing value changes between foreground background. Conventional coding techniques with transforms followed quantization result large artifacts along edges. To suppress these while preserving edges, we propose this...

We propose a deep autoencoder with graph topology inference and filtering to achieve compact representations of unorganized 3D point clouds in an unsupervised manner. Many previous works discretize points voxels then use lattice-based methods process learn spatial information; however, this leads inevitable discretization errors. In work, we handle raw without such compromise. The proposed networks follow the framework focus on designing decoder. encoder adopts similar architectures as...