1,720,976 research outputs found
Joint Learning of 3D Shape Retrieval and Deformation
No full textWe propose a novel technique for producing high-quality 3D models that match a given target object image or scan. Our method is based on retrieving an existing shape from a database of 3D models and then deforming its parts to match the target shape. Unlike previous approaches that independently focus on either shape retrieval or deformation, we propose a joint learning procedure that simultaneously trains the neural deformation module along with the embedding space used by the retrieval module. This enables our network to learn a deformation-aware embedding space, so that retrieved models are more amenable to match the target after an appropriate deformation. In fact, we use the embedding space to guide the shape pairs used to train the deformation module, so that it invests its capacity in learning deformations between meaningful shape pairs. Furthermore, our novel part-aware deformation module can work with inconsistent and diverse part-structures on the source shapes. We demonstrate the benefits of our joint training not only on our novel framework, but also on other state-of-the-art neural deformation modules proposed in recent years. Lastly, we also show that our jointly-trained method outperforms various non-joint baselines
View-dependent deformation fields for 2D editing of 3D models
No full textLe domaine de l’infographie s’est historiquement concentré sur le développement de modèles 3D statiques visant à représenter de manière réaliste des objets du monde réel. Dans ce mémoire, nous présentons une méthode permettant de créer des objets 3D non réalistes, représentés soit par des 3D Gaussian Splats, soit par des maillages, qui s’adaptent à des modifications 2D réalisées depuis des points de vue spécifiques. À partir d’un objet 3D donné, l’utilisateur sélectionne plusieurs points de vue et effectue des déformations interactives dans le plan image 2D associé à chaque vue. Notre méthode calcule alors un champ de déformation qui interpole ces modifications de manière fluide lorsque le point de vue change. L’idée centrale de notre approche est que ces déformations 2D ne nécessitent pas d’être liées à une même géométrie sous-jacente ni de partager un espace de déformation commun. Cette observation permet la construction de déformations dépendantes du point de vue, rendue possible grâce à plusieurs contributions techniques. Tout d’abord, nous introduisons une stratégie de composition permettant de combiner les déformations 2D après leur projection en 3D, de façon hiérarchique, à la manière de calques successifs dans un logiciel de retouche d’image. Nous proposons ensuite une méthode pour appliquer ces déformations 3D directement aux représentations en Gaussian Splats. Enfin, nous développons une interface intuitive pour la création des déformations 2D, reposant sur la manipulation d’un maillage 2D épousant l’image rendue de l’objet. Nous démontrons la richesse et la flexibilité de notre approche à travers divers cas d’usage, tels que l’ajout d’effets de style cartoon, la modification de personnages humains, l’adaptation de modèles 3D à des croquis ou caricatures 2D, la gestion des occlusions, ainsi que la réinterprétation de peintures non réalistes classiques sous forme de modèles 3D dynamiques.The field of computer graphics has been traditionally centered around the development of static 3D models designed to realistically represent objects from the physical world. In this thesis, we present a method for authoring non-realistic 3D objects, represented as either 3D Gaussian Splats or Meshes, that comply with 2D user edits from specific viewpoints. Given a 3D object, the user selects multiple viewpoints and interactively performs 2D deformations in the image plane of each view. Our method then computes a deformation field, a smooth interpolation between these 2D deformations as the viewpoint changes. A central insight of our work is that these 2D deformations do not need to be tied to a common underlying geometry or share a unified deformation space. This observation enables the construction of view-dependent deformations through several key contributions. First, we introduce a compositional blending strategy that allows 2D deformations, once lifted to 3D, to be combined in a layered manner akin to traditional image editing software. Second, we develop a method for applying these 3D deformations directly to 3D Gaussian Splats. Third, we propose an intuitive interface for authoring 2D deformations by editing a mesh that encapsulates a rendered image of the object. We demonstrate the expressiveness and versatility of our approach across a range of applications, including the stylization of objects with cartoon-like effects, the modification of human characters, the adaptation of 3D models to 2D sketches and caricatures, the resolution of occlusions, and the reinterpretation of classic non-realistic artworks as dynamic 3D models
Feature-Preserving Neural Surface Reconstruction Using the Dirichlet Energy of the Gauss Map
No full textGeometry acquisition from the real world is a fundamental but unsolved problem in computer graphics. With the evolution of sensors and new demands from modern applications, classical solutions are confronted with new challenges such as increased data volume and the necessity for real-time processing. Many of the new applications attempt to recover geometry from an RGB video instead of geometric data. Often, these are incorporated within real-time systems, where the efficiency of both time and memory are of crucial importance. Learning-based solutions are one type of approach that has received increasing attention.
However, neural networks tend to provide either over-smoothed results or they hallucinate geometry when there is not sufficient information. In this thesis, we analyze the low quality reconstructions and provide solutions for addressing this issue from two perspectives: gradient-domain processing of the implicit function and incorporation of total curvature as a weight.
In considering the surface as an implicit function, we express the loss function of the neural network in the gradient-domain. The advantage of using a loss term formulated in terms of gradients is two-fold. First, our method only requires self-supervised learning, removing the need for human labelers. Second, when learning the parameters of the network, the loss function gives more importance to the preservation of high-frequency details.
In considering the surface as a manifold, we filter an input point cloud (collected by the sensor, or sampled from a surface mesh) with respect to total curvature in a pre-processing step, using feature-preserving simplification to keep more points at highly curved regions. We introduce a new tool to estimate discrete total curvature for both point clouds and triangle meshes. Unlike existing approaches for discrete curvature estimation, our method bypasses the complex task of estimating the shape operator.
Our approach for total curvature estimation only requires the estimation of normals and a way to compute the Dirichlet energy – both well-studied tasks in geometry processing. Our approach demonstrates enhanced precision in estimating total curvature compared to classical geometry processing algorithms still in use today, as exemplified by those implemented in widely recognized libraries. Three applications making use of the estimated total curvature are demonstrated: mesh decimation, point cloud simplification, and feature-weighted surface reconstruction.
Numerous experiments are conducted to validate our framework. These include training from scratch and testing on the ScanNet benchmark; as well as applying the trained model to data collected with an iPhone. Both quantitative and qualitative results reveal the competitiveness of our approach with state-of-the-art methods.
In summary, this thesis presents two contributions. The first provides a new tool for computing total curvature. The second is a framework that improves the quality of surfaces reconstructed from a sequence of RGB images
Feature-Preserving Neural Surface Reconstruction Using the Dirichlet Energy of the Gauss Map
No full textGeometry acquisition from the real world is a fundamental but unsolved problem in computer graphics. With the evolution of sensors and new demands from modern applications, classical solutions are confronted with new challenges such as increased data volume and the necessity for real-time processing. Many of the new applications attempt to recover geometry from an RGB video instead of geometric data. Often, these are incorporated within real-time systems, where the efficiency of both time and memory are of crucial importance. Learning-based solutions are one type of approach that has received increasing attention.
However, neural networks tend to provide either over-smoothed results or they hallucinate geometry when there is not sufficient information. In this thesis, we analyze the low quality reconstructions and provide solutions for addressing this issue from two perspectives: gradient-domain processing of the implicit function and incorporation of total curvature as a weight.
In considering the surface as an implicit function, we express the loss function of the neural network in the gradient-domain. The advantage of using a loss term formulated in terms of gradients is two-fold. First, our method only requires self-supervised learning, removing the need for human labelers. Second, when learning the parameters of the network, the loss function gives more importance to the preservation of high-frequency details.
In considering the surface as a manifold, we filter an input point cloud (collected by the sensor, or sampled from a surface mesh) with respect to total curvature in a pre-processing step, using feature-preserving simplification to keep more points at highly curved regions. We introduce a new tool to estimate discrete total curvature for both point clouds and triangle meshes. Unlike existing approaches for discrete curvature estimation, our method bypasses the complex task of estimating the shape operator.
Our approach for total curvature estimation only requires the estimation of normals and a way to compute the Dirichlet energy – both well-studied tasks in geometry processing. Our approach demonstrates enhanced precision in estimating total curvature compared to classical geometry processing algorithms still in use today, as exemplified by those implemented in widely recognized libraries. Three applications making use of the estimated total curvature are demonstrated: mesh decimation, point cloud simplification, and feature-weighted surface reconstruction.
Numerous experiments are conducted to validate our framework. These include training from scratch and testing on the ScanNet benchmark; as well as applying the trained model to data collected with an iPhone. Both quantitative and qualitative results reveal the competitiveness of our approach with state-of-the-art methods.
In summary, this thesis presents two contributions. The first provides a new tool for computing total curvature. The second is a framework that improves the quality of surfaces reconstructed from a sequence of RGB images
View-Dependent Deformation Fields for 2D Editing of 3D Models
No full textWe propose a method for authoring non-realistic 3D objects (represented as either 3D Gaussian Splats or meshes), that comply with 2D edits from specific viewpoints. Namely, given a 3D object, a user chooses different viewpoints and interactively deforms the object in the 2D image plane of each view. The method then produces a ''deformation field'' - an interpolation between those 2D deformations in a smooth manner as the viewpoint changes. Our core observation is that the 2D deformations do not need to be tied to an underlying object, nor share the same deformation space. We use this observation to devise a method for authoring view-dependent deformations, holding several technical contributions: first, a novel way to compositionality-blend between the 2D deformations after lifting them to 3D - this enables the user to ''stack'' the deformations similarly to layers in an editing software, each deformation operating on the results of the previous; second, a novel method to apply the 3D deformation to 3D Gaussian Splats; third, an approach to author the 2D deformations, by deforming a 2D mesh encapsulating a rendered image of the object. We show the versatility and efficacy of our method by adding cartoonish effects to objects, providing means to modify human characters, fitting 3D models to given 2D sketches and caricatures, resolving occlusions, and recreating classic non-realistic paintings as 3D models.ACM/EG Expressive Symposium - WICED: Eurographics Workshop on Intelligent Cinematography and EditingUser Guided Creative Modelin
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Variations on the Author
Full text link“Variations on the Author” discusses two of Eduardo Coutinho’s recent films (Um Dia na Vida, from 2010, and Últimas Conversas, posthumously released in 2015) and their contribution to the general question of documentary authorship. The director’s filmography is characterized by a consistent yet self-effacing form of authorial self-inscription: Coutinho often features as an interviewer that rather than express opinions propels discourses; an interviewer that is good at listening. This mode of self-inscription characterizes him as an author who is not expressive but who is nonetheless markedly present on the screen. In Um Dia na Vida, however, Coutinho is completely absent form the image, while Últimas Conversas, on the contrary, includes a confessional prologue that moves the director from the margins to the center of his films. This article examines the ways in which these works stand out in the filmography of a director who offers new insights into the notion of cinematic authorship
Appropriate Similarity Measures for Author Cocitation Analysis
Full text linkWe provide a number of new insights into the methodological discussion about author cocitation analysis. We first argue that the use of the Pearson correlation for measuring the similarity between authors’ cocitation profiles is not very satisfactory. We then discuss what kind of similarity measures may be used as an alternative to the Pearson correlation. We consider three similarity measures in particular. One is the well-known cosine. The other two similarity measures have not been used before in the bibliometric literature. Finally, we show by means of an example that our findings have a high practical relevance.information science;Pearson correlation;cosine;similarity measure;author cocitation analysis
Dispelling the Myths Behind First-author Citation Counts
Full text linkWe conducted a full-scale evaluative citation analysis study of scholars in the XML research field to explore just how different from each other author rankings resulting from different citation counting methods actually are, and to demonstrate the capability of emerging data and tools on the Web in supporting more realistic citation counting methods. Our results contest some common arguments for the continued
use of first-author citation counts in the evaluation of scholars, such as high correlations between author rankings by first-author citation counts and other citation
counting methods, and high costs of using more realistic citation counting methods that are not well-supported by the ISI databases. It is argued that increasingly available digital full text research papers make it possible for citation analysis studies to go beyond what the ISI databases have directly supported and to employ more
sophisticated methods
- …
