- Advanced Vision and Imaging
- 3D Shape Modeling and Analysis
- Face recognition and analysis
- Computer Graphics and Visualization Techniques
- Human Motion and Animation
- Generative Adversarial Networks and Image Synthesis
- Optical measurement and interference techniques
- Facial Rejuvenation and Surgery Techniques
- Advanced Image Processing Techniques
- Human Pose and Action Recognition
- Video Surveillance and Tracking Methods
- Advanced Numerical Analysis Techniques
- Gaze Tracking and Assistive Technology
- Retinal Imaging and Analysis
- Robotics and Sensor-Based Localization
- Biometric Identification and Security
- Speech and Audio Processing
- Image Enhancement Techniques
- 3D Surveying and Cultural Heritage
- Hand Gesture Recognition Systems
- Advanced Image and Video Retrieval Techniques
- Image Processing Techniques and Applications
- Advanced Optical Imaging Technologies
- Medical Imaging Techniques and Applications
- Face and Expression Recognition
Walt Disney (Switzerland)
2012-2024
Walt Disney (United States)
2014-2023
National Tsing Hua University
2018
University of British Columbia
2006-2010
Carleton University
2005-2007
Image thresholding is a common task in many computer vision and graphics applications. The goal of an image to classify pixels as either "dark" or "light." Adaptive form that takes into account spatial variations illumination. We present technique for real-time adaptive using the integral input. Our extension previous method. However, our solution more robust illumination changes image. Additionally, method simple easy implement. suitable processing live video streams at frame-rate, making...
We present a new technique for passive and markerless facial performance capture based on anchor frames. Our method starts with high resolution per-frame geometry acquisition using state-of-the-art stereo reconstruction, proceeds to establish single triangle mesh that is propagated through the entire performance. Leveraging fact performances often contain repetitive subsequences, we identify frames as those which similar expressions manually chosen reference expression. Anchor are...
We present the first real-time high-fidelity facial capture method. The core idea is to enhance a global face tracker, which provides low-resolution mesh, with local regressors that add in medium-scale details, such as expression wrinkles. Our main observation although wrinkles appear different scales and at locations on face, they are locally very self-similar their visual appearance direct consequence of shape. therefore train from high-resolution data order predict geometry runtime....
We introduce a purely passive facial capture approach that uses only an array of video cameras, but requires no template geometry, special makeup or markers, and active lighting. obtain initial geometry using multi-view stereo, then use novel for automatically tracking texture detail across the frames. As result, we high-resolution sequence compatibly triangulated parameterized meshes. The resulting can be rendered with dynamically captured textures, while also consistently applying changes...
We present a new technique for passive and markerless facial performance capture based on anchor frames . Our method starts with high resolution per-frame geometry acquisition using state-of-the-art stereo reconstruction, proceeds to establish single triangle mesh that is propagated through the entire performance. Leveraging fact performances often contain repetitive subsequences, we identify as those which similar expressions manually chosen reference expression. Anchor are automatically...
A lot of research has recently focused on the problem capturing geometry and motion garments. Such work usually relies special markers printed fabric to establish temporally coherent correspondences between points garment's surface at different times. Unfortunately, this approach is tedious prevents capture off-the-shelf clothing made from interesting fabrics. In paper, we describe a marker-free garment that avoids these downsides. We parameterizations incomplete geometries extract each...
Fluid simulation is one of the most active research areas in computer graphics. However, it remains difficult to obtain measurements real fluid flows for validation simulated data. In this paper, we take a step direction capturing flow data such purposes. Specifically, present first time-resolved Schlieren tomography system full 3D, non-stationary gas on dense volumetric grid. uses 2D ray deflection reconstruct time-varying grid 3D refractive index values, which directly correspond physical...
This paper presents a new algorithm for multi-view reconstruction that demonstrates both accuracy and efficiency. Our method is based on robust binocular stereo matching, followed by adaptive point-based filtering of the merged point clouds, efficient, high-quality mesh generation. All aspects our are designed to be highly scalable with number views. technique produces most accurate results among current algorithms sparse viewpoints according Middlebury datasets. Additionally, we prove...
Abstract Progress in cloth simulation for computer animation and apparel design has led to a multitude of deformation models, each with its own way relating geometry, deformation, forces. As simulators improve, differences between these models become more important, but it is difficult choose model set parameters match given real material simply by looking at results. This paper provides measurement fitting methods that allow nonlinear be fit the observed particular sample. Unlike standard...
Recent research work has developed powerful generative models (e.g., StyleGAN2) that can synthesize complete human head images with impressive photorealism, enabling applications such as photorealistically editing real photographs. While these be trained on large collections of unposed images, their lack explicit 3D knowledge makes it difficult to achieve even basic control over viewpoint without unintentionally altering identity. On the other hand, recent Neural Radiance Field (NeRF)...
We present a new anatomically-constrained local face model and fitting approach for tracking 3D faces from 2D motion data in very high quality. In contrast to traditional global models, often built large set of blendshapes, we propose deformation composed many small subspaces spatially distributed over the face. Our offers far more flexibility expressiveness than blendshape even with much smaller size. This would typically come at cost reduced robustness, particular during under-constrained...
Even though the human eye is one of central features individual appearance, its shape has so far been mostly approximated in our community with gross simplifications. In this paper we demonstrate that there a lot individuality to every eye, fact common practices for 3D generation do not consider. To faithfully reproduce all intricacies propose novel capture system capable accurately reconstructing visible parts eye: white sclera , transparent cornea and non-rigidly deforming colored iris ....
In recent years, sophisticated image-based reconstruction methods for the human face have been developed. These capture highly detailed static and dynamic geometry of whole face, or specific models regions, such as hair, eyes eye lids. Unfortunately, to mouth cavity in general, teeth particular, received very little attention. The accurate rendering teeth, however, is crucial realistic display facial expressions, currently high quality animations resort tooth row created by tedious manual...
We present a method to acquire dynamic properties of facial skin appearance, including diffuse albedo encoding blood flow, specular intensity, and per-frame high resolution normal maps for performance sequence. The reconstructs these from purely passive multi-camera setup, without the need polarization or requiring temporally multiplexed illumination. Hence, it is very well suited integration with existing systems capture. To solve this seemingly underconstrained problem, we demonstrate that...
Facial landmark detection is a fundamental task for many consumer and high-end applications almost entirely solved by machine learning methods today. Existing datasets used to train such algorithms are primarily made up of only low resolution images, current limited inputs comparable quality as the training dataset. On other hand, high imagery becoming increasingly more common cameras improve in every year. Therefore, there need that can leverage rich information available imagery. Naively...
We propose a new light-weight face capture system capable of reconstructing both high-quality geometry and detailed appearance maps from single exposure. Unlike currently employed acquisition systems, the proposed technology does not require active illumination hence can readily be integrated with passive photogrammetry solutions. These solutions are in widespread use for 3D scanning humans as they assembled off-the-shelf hardware components, but lack capability estimating appearance. This...
Style transfer between images is an artistic application of CNNs, where the ‘style’ one image transferred onto another while preserving latter’s content. The state art in neural style based on Adaptive Instance Normalization (AdaIN), a technique that transfers statistical properties features to content image, and can large number styles real time. However, AdaIN global operation; thus local geometric structures are often ignored during transfer. We propose Convolutions (AdaConv), generic...
Flora is an element in many computer-generated scenes. But trees, bushes and plants have complex geometry appearance, are difficult to model manually. One way address this capture models directly from the real world. Existing techniques focused on extracting macro structure such as branching of or broad-leaved with a relatively small number surfaces. This paper presents finer scale technique demonstrate for first time processing densely leaved foliage - computation 3D structure, plus...
Facial scanning has become ubiquitous in digital media, but so far most efforts have focused on reconstructing the skin. Eye reconstruction, other hand, received only little attention, and current state-of-the-art method is cumbersome for actor, time-consuming, requires carefully setup calibrated hardware. These constraints currently make eye capture impractical general use. We present first approach high-quality lightweight capture, which leverages a database of pre-captured eyes to guide...
Abstract Facial appearance capture is now firmly established within academic research and used extensively across various application domains, perhaps most prominently in the entertainment industry through design of virtual characters video games films. While significant progress has occurred over last two decades, no single survey currently exists that discusses similarities, differences, practical considerations available techniques as applied to human faces. A central difficulty facial...
2D portrait animation has experienced significant advancements in recent years. Much research utilized the prior knowledge embedded large generative diffusion models to enhance high-quality image manipulation. However, most methods only focus on generating RGB images as output, and co-generation of consistent visual plus 3D output remains largely under-explored. In our work, we propose jointly learn appearance depth simultaneously a diffusion-based generator. Our method embraces end-to-end...
The scanning of 3D geometry has become a popular way capturing the shape real-world objects. Transparent objects, however, pose problems for traditional methods. We present visible light tomographic reconstruction method recovering transparent such as glass. Our setup is relatively simple to implement, and accounts refraction, which can be significant problem in tomography.
Abstract The presence of characteristic fine folds is important for modeling realistic looking virtual garments. While recent garment capture techniques are quite successful at capturing the low‐frequency shape and motion over time, they often fail to numerous high‐frequency folds, reducing realism reconstructed space‐time models. In our work we propose a method reintroducing into captured models using data‐driven dynamic wrinkling. We first estimate position based on original video footage...
We introduce a purely passive facial capture approach that uses only an array of video cameras, but requires no template geometry, special makeup or markers, and active lighting. obtain initial geometry using multi-view stereo, then use novel for automatically tracking texture detail across the frames. As result, we high-resolution sequence compatibly triangulated parameterized meshes. The resulting can be rendered with dynamically captured textures, while also consistently applying changes...