A5053518240- 3D Shape Modeling and Analysis
- Image Retrieval and Classification Techniques
- Advanced Image and Video Retrieval Techniques
- Computer Graphics and Visualization Techniques
- Medical Image Segmentation Techniques
- Human Pose and Action Recognition
- Advanced Vision and Imaging
- Image Processing and 3D Reconstruction
- Generative Adversarial Networks and Image Synthesis
- Multimodal Machine Learning Applications
- 3D Surveying and Cultural Heritage
- Additive Manufacturing and 3D Printing Technologies
- Speech and Audio Processing
- Music Technology and Sound Studies
- Interactive and Immersive Displays
- Music and Audio Processing
- Architecture and Computational Design
- Optimization and Packing Problems
- Face recognition and analysis
- Video Analysis and Summarization
- Optimization and Search Problems
- Modular Robots and Swarm Intelligence
- Caching and Content Delivery
- Domain Adaptation and Few-Shot Learning
- Computational Geometry and Mesh Generation
Centre National de la Recherche Scientifique
2017-2018
DNEG (United Kingdom)
2018
Laboratoire d'Informatique de l'École Polytechnique
2017-2018
Tel Aviv University
2011-2018
École Polytechnique
2017
Academic College of Tel Aviv-Yafo
2007
Abstract In this paper, we propose PCPN ET , a deep‐learning based approach for estimating local 3D shape properties in point clouds. contrast to the majority of prior techniques that concentrate on global or mid‐level attributes, e.g., classification semantic labeling, suggest patch‐based learning method, which series patches at multiple scales around each is encoded structured manner. Our especially well‐adapted such as normals (both unoriented and oriented) curvature from raw clouds...
We introduce an algorithm for unsupervised co-segmentation of a set shapes so as to reveal the semantic shape parts and establish their correspondence across set. The input may exhibit significant variability where do not admit proper spatial alignment corresponding in any pair be geometrically dissimilar. Our can handle such challenging sets since, first, we perform co-analysis descriptor space, combination descriptors relates independently pose, location, cardinality. Secondly, exploit key...
We introduce an algorithm for unsupervised co-segmentation of a set shapes so as to reveal the semantic shape parts and establish their correspondence across set. The input may exhibit significant variability where do not admit proper spatial alignment corresponding in any pair be geometrically dissimilar. Our can handle such challenging sets since, first, we perform co-analysis descriptor space , combination descriptors relates independently pose, location, cardinality. Secondly, exploit...
We present a shape segmentation method for complete and incomplete shapes. The key idea is to directly optimize the decomposition based on characterization of expected geometry part in shape. Rather than setting number parts advance, we search smallest that admit geometric parts. an intermediate-level analysis, where first decomposed into approximate convex components, which are then merged consistent nonlocal signature. Our designed handle shapes, represented by point clouds. show results...
We present a technique to synthesize time-varying weathered textures. Given single texture image as input, the degree of weathering at different regions input is estimated by prevalence analysis patches. This information then allows gracefully increase or decrease popularity patches, simulating evolution appearance both backward and forward in time. Our method can be applied wide variety textures since reaction material effects physically-oblivious learned from itself. The process evolves...
Computing similarities or distances between 3D shapes is a crucial building block for numerous tasks, including shape retrieval, exploration and classification. Current state-of-the-art distance measures mostly consider the overall appearance of are less sensitive to fine changes in structure geometry. We present edit (SHED) that amount effort needed transform one into other, terms re-arranging parts match other shape, as well possibly adding removing parts. The takes account both similarity...
We present a novel system for browsing through very large set of images according to similarity. The are dynamically placed on 2D canvas next their nearest neighbors in high-dimensional feature space. layout and choice is generated on-the-fly during user interaction, reflecting the user's navigation tendencies interests. This intuitive solution image provides continuous experience navigating an infinite grid arranged by In contrast common multidimensional embedding methods, our does not...
Abstract We present a robust method to find region‐level correspondences between shapes, which are invariant changes in geometry and applicable across multiple shape representations. generate simplified graphs by jointly decomposing the devise an adapted graph‐matching technique, from we infer regions. The designed primarily capture overall structure of without reflecting precise information about each region, enables us shapes that might have significant geometric differences. Moreover, due...
Abstract Large datasets of 3D objects require an intuitive way to browse and quickly explore shapes from the collection. We present a dynamic map where similar are placed next each other. Similarity between models exists in high dimensional space which cannot be accurately expressed two map. solve this discrepancy by providing local with pan capabilities user interface that resembles online experience navigating through geographical maps. As navigates map, new appear correspond specific...
We introduce an algorithm for unsupervised co-segmentation of a set shapes so as to reveal the semantic shape parts and establish their correspondence across set. The input may exhibit significant variability where do not admit proper spatial alignment corresponding in any pair be geometrically dissimilar. Our can handle such challenging sets since, first, we perform co-analysis descriptor space, combination descriptors relates independently pose, location, cardinality. Secondly, exploit key...
Abstract In this paper we introduce a video post-processing method that enhances the rhythm of dancing performance, in sense movements are more time to beat music. The performance as observed is analyzed and segmented into motion intervals delimited by beats. We present an image-space extract beats detecting frames at which there significant change direction or stops. then synchronized with music such many possible matched little time-warping distortion video. show two applications for...
Region-based correspondence (RBC) is a highly relevant and non-trivial computer vision problem. Given two 3D shapes, RBC seeks segments/regions on these shapes that can be reliably put in correspondence. The problem thus consists both finding the regions determining correspondences between them. This statement similar to of "biclustering ", implying cast as biclustering Here, we exploit this implication by tackling via novel approach, called S4B (spatially smooth spike slab biclustering),...
We introduce Replay, a collection of multi-view, multi-modal videos humans interacting socially. Each scene is filmed in high production quality, from different view-points with several static cameras, as well wearable action and recorded large array microphones at positions the room. Overall, dataset contains over 4000 minutes footage 7 million timestamped high-resolution frames annotated camera poses partially foreground masks. The Replay has many potential applications, such novel-view...
We present Meta 3D AssetGen (AssetGen), a significant advancement in text-to-3D generation which produces faithful, high-quality meshes with texture and material control. Compared to works that bake shading the object's appearance, outputs physically-based rendering (PBR) materials, supporting realistic relighting. generates first several views of object factored shaded albedo appearance channels, then reconstructs colours, metalness roughness 3D, using deferred loss for efficient...
The recent availability and adaptability of text-to-image models has sparked a new era in many related domains that benefit from the learned text priors as well high-quality fast generation capabilities, one which is texture for 3D objects. Although methods achieve impressive results by using networks, combination global consistency, quality, speed, crucial advancing to real-world applications, remains elusive. To end, we introduce Meta TextureGen: feedforward method comprised two sequential...
We introduce Meta 3D Gen (3DGen), a new state-of-the-art, fast pipeline for text-to-3D asset generation. 3DGen offers creation with high prompt fidelity and high-quality shapes textures in under minute. It supports physically-based rendering (PBR), necessary relighting real-world applications. Additionally, generative retexturing of previously generated (or artist-created) using additional textual inputs provided by the user. integrates key technical components, AssetGen TextureGen, that we...
Machine learning techniques are not often associated with artistic work such as visual effects production. Nevertheless, these can save a lot of time for artists when used in the right context. In recent years, deep have become widely tool powerful frameworks that be employed production environment. We present two solutions were integrated into our pipeline and current productions. One method generates high quality images from compressed video file contains various compression artifacts. The...
We introduce Replay, a collection of multi-view, multi-modal videos humans interacting socially. Each scene is filmed in high production quality, from different viewpoints with several static cameras, as well wearable action and recorded large array microphones at positions the room. Overall, dataset contains over 4000 minutes footage 7 million timestamped high-resolution frames annotated camera poses partially foreground masks. The Replay has many potential applications, such novel-view...
Understanding semantic similarity among images is the core of a wide range computer vision applications. An important step towards this goal to collect and learn human perceptions. Interestingly, context often ambiguous as can be perceived with emphasis on different aspects, which may contradictory each other. In paper, we present method for learning images, inferring their latent aspects embedding them into multi-spaces corresponding aspects. We consider multi-embedding problem an...
We present a robust method to find region-level correspondences between shapes, which are invariant changes in geometry and applicable across multiple shape representations. generate simplified graphs by jointly decomposing the devise an adapted graph-matching technique, from we infer regions. The designed primarily capture overall structure of without reflecting precise information about each region, enables us shapes that might have significant geometric differences. Moreover, due special...
Understanding semantic similarity among images is the core of a wide range computer graphics and vision applications. However, visual context often ambiguous as that can be perceived with emphasis on different attributes. In this paper, we present method for learning images, inferring their latent attributes embedding them into multi-spaces corresponding to each attribute. We consider multi-embedding problem an optimization function evaluates embedded distances respect qualitative...