3DOR 19

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/2632755

Browse

Now showing 1 - 8 of 8

CMH: Coordinates Manifold Harmonics for Functional Remeshing
(The Eurographics Association, 2019) Marin, Riccardo; Melzi, Simone; Musoni, Pietro; Bardon, Filippo; Tarini, Marco; Castellani, Umberto; Biasotti, Silvia and Lavoué, Guillaume and Veltkamp, Remco
In digital world reconstruction, 2-dimensional surface of real objects are often obtained as polygonal meshes after an acquisition procedure using 3D sensors. However, such representation requires several manual efforts from highly experts to correct the irregularity of tessellation and make it suitable for professional applications, such as those in the gaming or movie industry. Moreover, for modelling and animation purposes it is often required that the same connectivity is shared among two or more different shapes. In this paper we propose a new method that exploits a remeshing-by-matching approach where the observed noisy shape inherits a regular tessellation from a target shape which already satisfies the professional constraints. A fully automatic pipeline is introduced based on a variation of the functional mapping framework. In particular, a new set of basis functions, namely the Coordinates Manifold Harmonics (CMH), is properly designed for this tessellation transfer task. In our experiments an exhaustive quantitative and quality evaluation is reported for human body shapes in T-pose where the effectiveness of the proposed functional remeshing is clearly shown in comparison with other methods.
Depth-Based Face Recognition by Learning from 3D-LBP Images
(The Eurographics Association, 2019) Neto, Joao Baptista Cardia; Marana, Aparecido Nilceu; Ferrari, Claudio; Berretti, Stefano; Bimbo, Alberto Del; Biasotti, Silvia and Lavoué, Guillaume and Veltkamp, Remco
In this paper, we propose a hybrid framework for face recognition from depth images, which is both effective and efficient. It consists of two main stages: First, the 3DLBP operator is applied to the raw depth data of the face, and used to build the corresponding descriptor images (DIs). However, such operator quantizes relative depth differences over/under +-7 to the same bin, so as to generate a fixed dimensional descriptor. To account for this behavior, we also propose a modification of the traditional operator that encodes depth differences using a sigmoid function. Then, a not-so-deep (shallow) convolutional neural network (SCNN) has been designed that learns from the DIs. This architecture showed two main advantages over the direct application of deep-CNN (DCNN) to depth images of the face: On the one hand, the DIs are capable of enriching the raw depth data, emphasizing relevant traits of the face, while reducing their acquisition noise. This resulted decisive in improving the learning capability of the network; On the other, the DIs capture low-level features of the face, thus playing the role for the SCNN as the first layers do in a DCNN architecture. In this way, the SCNN we have designed has much less layers and can be trained more easily and faster. Extensive experiments on low- and high-resolution depth face datasets confirmed us the above advantages, showing results that are comparable or superior to the state-of-the-art, using by far less training data, time, and memory occupancy of the network.
Feature Curve Extraction on Triangle Meshes
(The Eurographics Association, 2019) Moscoso Thompson, Elia; Arvanitis, G.; Moustakas, Konstantinos; Hoang-Xuan, N.; Nguyen, E. R.; Tran, M.; Lejemble, T.; Barthe, L.; Mellado, N.; Romanengo, C.; Biasotti, S.; FALCIDIENO, BIANCA; Biasotti, Silvia and Lavoué, Guillaume and Veltkamp, Remco
This paper presents the results of the SHREC'19 track: Feature curve extraction on triangle meshes. Given a model, the challenge consists in automatically extracting a subset of the mesh vertices that jointly represent a feature curve. As an optional task, participants were requested to send also a similarity evaluation among the feature curves extracted. The various approaches presented by the participants are discussed, together with their results. The proposed methods highlight different points of view of the problem of feature curve extraction. It is interesting to see that it is possible to deal with this problem with good results, despite the different approaches.
Generalizing Discrete Convolutions for Unstructured Point Clouds
(The Eurographics Association, 2019) Boulch, Alexandre; Biasotti, Silvia and Lavoué, Guillaume and Veltkamp, Remco
Point clouds are unstructured and unordered data, as opposed to images. Thus, most of machine learning approaches, developed for images, cannot be directly transferred to point clouds. It usually requires data transformation such as voxelization, inducing a possible loss of information. In this paper, we propose a generalization of the discrete convolutional neural networks (CNNs) able to deal with sparse input point cloud. We replace the discrete kernels by continuous ones. The formulation is simple, does not set the input point cloud size and can easily be used for neural network design similarly to 2D CNNs. We present experimental results, competitive with the state of the art, on shape classification, part segmentation and semantic segmentation for large scale clouds.
Matching Humans with Different Connectivity
(The Eurographics Association, 2019) Melzi, S.; Marin, R.; Rodolà, E.; Castellani, U.; Ren, J.; Poulenard, A.; Wonka, P.; Ovsjanikov, M.; Biasotti, Silvia and Lavoué, Guillaume and Veltkamp, Remco
Objects Matching is a ubiquitous problem in computer science with particular relevance for many applications; property transfer between 3D models and statistical study for learning are just some remarkable examples. The research community spent a lot of effort to address this problem, and a large and increased set of innovative methods has been proposed for its solution. In order to provide a fair comparison among these methods, different benchmarks have been proposed. However, all these benchmarks are domain specific, e.g., real scans coming from the same acquisition pipeline, or synthetic watertight meshes with the same triangulation. To the best of our knowledge, no cross-dataset comparisons have been proposed to date. This track provides the first matching evaluation in terms of large connectivity changes between models that come from totally different modeling methods. We provide a dataset of 44 shapes with dense correspondence as obtained by a highly accurate shape registration method (FARM). Our evaluation proves that connectivity changes lead to Objects Matching difficulties and we hope this will promote further research in matching shapes with wildly different connectivity.
mpLBP: An Extension of the Local Binary Pattern to Surfaces based on an Efficient Coding of the Point Neighbours
(The Eurographics Association, 2019) Moscoso Thompson, Elia; Biasotti, Silvia; Digne, Julie; Chaine, Raphaëlle; Biasotti, Silvia and Lavoué, Guillaume and Veltkamp, Remco
The description of surface textures in terms of repeated colorimetric and geometric local surface variations is a crucial task for several applications, such as object interpretation or style identification. Recently, methods based on extensions to the surface meshes of the Local Binary Pattern (LBP) or the Scale-Invariant Feature Transform (SIFT) descriptors have been proposed for geometric and colorimetric pattern retrieval and classification. With respect to the previous works, we consider a novel LBPbased descriptor based on the assignment of the point neighbours into sectors of equal area and a non-uniform, multiple ring sampling. Our method is able to deal with surfaces represented as point clouds. Experiments on different benchmarks confirm the competitiveness of the method within the existing literature, in terms of accuracy and computational complexity.
POP: Full Parametric model Estimation for Occluded People
(The Eurographics Association, 2019) Marin, Riccardo; Melzi, Simone; Mitra, Niloy J.; Castellani, Umberto; Biasotti, Silvia and Lavoué, Guillaume and Veltkamp, Remco
In the last decades, we have witnessed advances in both hardware and associated algorithms resulting in unprecedented access to volumes of 2D and, more recently, 3D data capturing human movement. We are no longer satisfied with recovering human pose as an image-space 2D skeleton, but seek to obtain a full 3D human body representation. The main challenges in acquiring 3D human shape from such raw measurements are identifying which parts of the data relate to body measurements and recovering from partial observations, often arising out of severe occlusion. For example, a person occluded by a piece of furniture, or being self-occluded in a profile view. In this paper, we propose POP, a novel and efficient paradigm for estimation and completion of human shape to produce a full parametric 3D model directly from single RGBD images, even under severe occlusion. At the heart of our method is a novel human body pose retrieval formulation that explicitly models and handles occlusion. The retrieved result is then refined by a robust optimization to yield a full representation of the human shape. We demonstrate our method on a range of challenging real world scenarios and produce high-quality results not possible by competing alternatives. The method opens up exciting AR/VR application possibilities by working on 'in-the-wild' measurements of human motion.
Shape Correspondence with Isometric and Non-Isometric Deformations
(The Eurographics Association, 2019) Dyke, R. M.; Stride, C.; Lai, Y.-K.; Rosin, P. L.; Aubry, M.; Boyarski, A.; Bronstein, A. M.; Bronstein, M. M.; Cremers, D.; Fisher, M.; Groueix, T.; Guo, D.; Kim, V. G.; Kimmel, R.; Lähner, Z.; Li, K.; Litany, O.; Remez, T.; Rodolà, E.; Russell, B. C.; Sahillioglu, Y.; Slossberg, R.; Tam, G. K. L.; Vestner, M.; Wu, Z.; Yang, J.; Biasotti, Silvia and Lavoué, Guillaume and Veltkamp, Remco
The registration of surfaces with non-rigid deformation, especially non-isometric deformations, is a challenging problem. When applying such techniques to real scans, the problem is compounded by topological and geometric inconsistencies between shapes. In this paper, we capture a benchmark dataset of scanned 3D shapes undergoing various controlled deformations (articulating, bending, stretching and topologically changing), along with ground truth correspondences. With the aid of this tiered benchmark of increasingly challenging real scans, we explore this problem and investigate how robust current state-of- the-art methods perform in different challenging registration and correspondence scenarios. We discover that changes in topology is a challenging problem for some methods and that machine learning-based approaches prove to be more capable of handling non-isometric deformations on shapes that are moderately similar to the training set.

Browse

Browsing 3DOR 19 by Subject "Computing methodologies"

Results Per Page

Sort Options