A5064927253- 3D Shape Modeling and Analysis
- Computer Graphics and Visualization Techniques
- Advanced Numerical Analysis Techniques
- Advanced Vision and Imaging
- Human Motion and Animation
- Computational Geometry and Mesh Generation
- Interactive and Immersive Displays
- Human Pose and Action Recognition
- 3D Surveying and Cultural Heritage
- Augmented Reality Applications
- Additive Manufacturing and 3D Printing Technologies
- Hand Gesture Recognition Systems
- Generative Adversarial Networks and Image Synthesis
- Advanced Image Processing Techniques
- Image Processing and 3D Reconstruction
- Image Enhancement Techniques
- Advanced Image and Video Retrieval Techniques
- Manufacturing Process and Optimization
- Tactile and Sensory Interactions
- Textile materials and evaluations
- Video Analysis and Summarization
- Image Processing Techniques and Applications
- Advanced Materials and Mechanics
- Force Microscopy Techniques and Applications
- Image Retrieval and Classification Techniques
ETH Zurich
2015-2024
Cognitive Neuroimaging Lab
2022
Human Computer Interaction (Switzerland)
2022
Courant Institute of Mathematical Sciences
2018
New York University
2018
Tel Aviv University
2018
Board of the Swiss Federal Institutes of Technology
2018
Czech Academy of Sciences, Institute of Computer Science
2018
Intel (United States)
2018
University of Washington
2018
This paper reviews the 2nd NTIRE challenge on single image super-resolution (restoration of rich details in a low resolution image) with focus proposed solutions and results. The had 4 tracks. Track 1 employed standard bicubic downscaling setup, while Tracks 2, 3 realistic unknown downgrading operators simulating camera acquisition pipeline. were learnable through provided pairs high train images. tracks 145, 114, 101, 113 registered participants, resp., 31 teams competed final testing...
Modern camera calibration and multiview stereo techniques enable users to smoothly navigate between different views of a scene captured using standard cameras. The underlying automatic 3D reconstruction methods work well for buildings regular structures but often fail on vegetation, vehicles, other complex geometry present in everyday urban scenes. Consequently, missing depth information makes Image-Based Rendering (IBR) such scenes very challenging. Our goal is provide plausible...
We present a novel approach to remesh surface into an isotropic triangular or quad-dominant mesh using unified local smoothing operator that optimizes both the edge orientations and vertex positions in output mesh. Our algorithm produces meshes with high isotropy while naturally aligning snapping edges sharp features. The method is simple implement parallelize, it can process variety of input representations, such as point clouds, range scans triangle meshes. full pipeline executes instantly...
Imbalance suggests a feeling of dynamism and movement in static objects. It is therefore not surprising that many 3D models stand impossibly balanced configurations. As long as the remain computer this no consequence: laws physics do apply. However, fabrication through printing breaks illusion: printed topple instead standing initially intended. We propose to assist users producing novel, properly designs by interactively deforming an existing model. formulate balance optimization energy...
We present a detail-driven deep neural network for point set upsampling. A high-resolution is essential point-based rendering and surface reconstruction. Inspired by the recent success of image super-resolution techniques, we progressively train cascade patch-based upsampling networks on different levels detail end-to-end. propose series architectural design contributions that lead to substantial performance boost. The effect each technical contribution demonstrated in an ablation study....
Recent deep learning approaches to single image superresolution have achieved impressive results in terms of traditional error measures and perceptual quality. However, each case it remains challenging achieve high quality for large upsampling factors. To this end, we propose a method (ProSR) that is progressive both architecture training: the network upsamples an intermediate steps, while process organized from easy hard, as done curriculum learning. obtain more photorealistic results,...
We introduce a novel deep learning framework for data-driven motion retargeting between skeletons, which may have different structure, yet corresponding to homeomorphic graphs. Importantly, our approach learns how retarget without requiring any explicit pairing the motions in training set. leverage fact that skeletons be reduced common primal skeleton by sequence of edge merging operations, we refer as skeletal pooling. Thus, main technical contribution is introduction differentiable...
Solid shapes in computer graphics are often represented with boundary descriptions, e.g. triangle meshes, but animation, physically-based simulation, and geometry processing more realistic accurate when explicit volume representations available. Tetrahedral meshes which exactly contain (interpolate) the input description desirable difficult to construct for a large class of meshes. Character CAD models composed many connected components numerous self-intersections, non-manifold pieces, open...
Spinning tops and yo-yos have long fascinated cultures around the world with their unexpected, graceful motions that seemingly elude gravity. We present an algorithm to generate designs for spinning objects by optimizing rotational dynamics properties. As input, user provides a solid 3D model desired axis of rotation. Our approach then modifies mass distribution such principal directions moment inertia align target rotation frame. augment creating voids inside its volume, interior fill...
We propose a stretch-sensing soft glove to interactively capture hand poses with high accuracy and without requiring an external optical setup. demonstrate how our device can be fabricated calibrated at low cost, using simple tools available in most fabrication labs. To reconstruct the pose from capacitive sensors embedded glove, we deep network architecture that exploits spatial layout of sensor itself. The is trained only once, inexpensive off-the-shelf reconstruction system gather...
We present a scalable approach for the optimization of flip-preventing energies in general context simplicial mappings and specifically mesh parameterization. Our iterative minimization is based on observation that many distortion can be optimized indirectly by minimizing family simpler proxy energies. Minimization these proxies natural extension local/global ARAP energy. algorithm simple to implement scales datasets with millions faces. demonstrate our computation maps minimize conformal or...
Abstract The mechanical wiring between cells and their surroundings is fundamental to the regulation of complex biological processes during tissue development, repair or pathology. Traction force microscopy (TFM) enables determination actuating forces. Despite progress, important limitations with intrusion effects in low resolution 2D pillar-based methods disruptive intermediate steps cell removal substrate relaxation high-resolution continuum TFM need be overcome. Here we introduce a novel...
Abstract Mappings and deformations are ubiquitous in geometry processing, shape modeling, animation. Numerous deformation energies have been proposed to tackle problems like mesh parameterization volumetric deformations. We present an algorithm that modifies any energy guarantee a locally injective mapping, i.e., without inverted elements. Our formulation can be used compute continuous planar or piecewise‐linear maps it uses barrier term prevent Differently from previous methods, we...
We present a scalable approach for the optimization of flip-preventing energies in general context simplicial mappings and specifically mesh parameterization. Our iterative minimization is based on observation that many distortion can be optimized indirectly by minimizing family simpler proxy energies. Minimization these proxies natural extension local/global ARAP energy. algorithm simple to implement scales datasets with millions faces. demonstrate our computation maps minimize conformal or...
We propose a novel learnable representation for detail preserving shape deformation. The goal of our method is to warp source match the general structure target shape, while surface details source. Our extends traditional cage-based deformation technique, where enclosed by coarse control mesh termed cage, and translations prescribed on cage vertices are interpolated any point via special weight functions. use this sparse scaffolding enables regardless shape's intricacy topology. key...
Abstract We present a method to automatically segment indoor scenes by detecting repeated objects. Our algorithm scales datasets with 198 million points and does not require any training data. propose trivially parallelizable preprocessing step, which compresses point cloud into collection of nearly‐planar patches related geometric transformations. This representation enables us robustly filter out noise greatly reduces the computational cost memory requirements our method, enabling...
Many algorithms on meshes require the minimization of composite objectives, i.e. , energies that are compositions simpler parts. Canonical examples include mesh parameterization and deformation. We propose a second order optimization approach exploits this structure to efficiently converge local minimum. Our main observation is convex-concave decomposition energy constituents simple readily available in many cases practical relevance graphics. utilize such decompositions define tight convex...
Existing deep learning approaches to single image super-resolution have achieved impressive results but mostly assume a setting with fixed pairs of high resolution and low images. However, robustly address realistic upscaling scenarios where the relation between images is unknown, blind required. To this end, we propose solution that relies on three components: First, use degradation aware SR network synthesize HR given corresponding blur kernel. Second, train kernel discriminator analyze...
Animating a newly designed character using motion capture (mocap) data is long standing problem in computer animation. A key consideration the skeletal structure that should correspond to available mocap data, and shape deformation joint regions, which often requires tailored, pose-specific refinement. In this work, we develop neural technique for articulating 3D characters enveloping with pre-defined produces high quality pose dependent deformations. Our framework learns rig skin same...
We present GANimator, a generative model that learns to synthesize novel motions from single, short motion sequence. GANimator generates resemble the core elements of original motion, while simultaneously synthesizing and diverse movements. Existing data-driven techniques for synthesis require large dataset which contains desired specific skeletal structure. By contrast, only requires training on single sequence, enabling variety structures e.g., bipeds, quadropeds, hexapeds, more. Our...
The emergence of neural networks has revolutionized the field motion synthesis. Yet, learning to unconditionally synthesize motions from a given distribution remains challenging, especially when are highly diverse. In this work, we present MoDi - generative model trained in an unsupervised setting extremely diverse, unstructured and unlabeled dataset. During inference, can high-quality, diverse motions. Despite lack any structure dataset, our yields well-behaved structured latent space,...
We present a complete design pipeline that allows non-expert users to and analyze masonry structures without any structural knowledge. optimize the force layouts both geometrically topologically, finding self-supported structure is as close possible given target surface. The generated are tessellated into hexagonal blocks with pattern prevents sliding failure. models can be used in physically plausible virtual environments or 3D printed assembled reinforcements.
We introduce a meta-representation that represents the essence of family shapes. The learns configurations shape parts are common across family, and encapsulates this knowledge with system geometric distributions encode relative arrangements parts. Thus, instead predefined priors, what characterizes is directly learned from set input constructed co-segmented shapes known correspondence. It can then be used in several applications where we seek to preserve identity as members family....
Self-supporting structures are prominent in historical and contemporary architecture due to advantageous structural properties efficient use of material. Computer graphics research has recently contributed new design tools that allow creating interactively exploring self-supporting freeform designs. However, the physical construction such remains challenging, even on small scales. Current processes require extensive formwork during assembly, which quickly leads prohibitively high costs for...
We consider the problem of reproducing look and details a 3D object on surface that is confined to given volume. Classic examples such "appearance-mimicking" surfaces are bas-reliefs: decorations artwork depicting recognizable scenes using only thin volumetric space. The design bas-reliefs has fascinated humankind for millennia it extensively used coins, medals, pottery other art forms. propose unified framework create depict certain shapes from prescribed viewpoints, as generalization...