Aaron Hertzmann

ORCID: 0000-0001-9667-0292
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Computer Graphics and Visualization Techniques
  • Advanced Vision and Imaging
  • 3D Shape Modeling and Analysis
  • Human Pose and Action Recognition
  • Human Motion and Animation
  • Aesthetic Perception and Analysis
  • Generative Adversarial Networks and Image Synthesis
  • Video Analysis and Summarization
  • Advanced Numerical Analysis Techniques
  • Visual Attention and Saliency Detection
  • Advanced Image and Video Retrieval Techniques
  • 3D Surveying and Cultural Heritage
  • Image Retrieval and Classification Techniques
  • Advanced Image Processing Techniques
  • Data Visualization and Analytics
  • Image Processing and 3D Reconstruction
  • Color perception and design
  • Virtual Reality Applications and Impacts
  • Music Technology and Sound Studies
  • Optical measurement and interference techniques
  • Gaussian Processes and Bayesian Inference
  • Robotic Locomotion and Control
  • Handwritten Text Recognition Techniques
  • Visual perception and processing mechanisms
  • Image Enhancement Techniques

Adobe Systems (United States)
2016-2025

New York University
1998-2023

University of Washington
2002-2023

Microsoft (United States)
2001-2023

University of Massachusetts Amherst
2021

University of Toronto
2008-2019

University of California, Berkeley
2014

Camera shake during exposure leads to objectionable image blur and ruins many photographs. Conventional blind deconvolution methods typically assume frequency-domain constraints on images, or overly simplified parametric forms for the motion path camera shake. Real motions can follow convoluted paths, a spatial domain prior better maintain visually salient characteristics. We introduce method remove effects of from seriously blurred images. The assumes uniform over negligible in-plane...

10.1145/1141911.1141956 article EN ACM Transactions on Graphics 2006-07-01

This paper describes a new framework for processing images by example, called “image analogies.” The involves two stages: design phase, in which pair of images, with one image purported to be “filtered” version the other, is presented as “training data”; and an application learned filter applied some target order create “analogous” filtered result. Image analogies are based on simple multi-scale autoregression, inspired primarily recent results texture synthesis. By choosing different types...

10.1145/383259.383295 article EN 2001-08-01

We introduce Gaussian process dynamical models (GPDM) for nonlinear time series analysis, with applications to learning of human pose and motion from high-dimensionalmotion capture data. A GPDM is a latent variable model. It comprises low-dimensional space associated dynamics, map the an observation space. marginalize out model parameters in closed-form, using priors both dynamics mappings. This results non-parametric systems that accounts uncertainty demonstrate approach, compare four...

10.1109/tpami.2007.1167 article EN IEEE Transactions on Pattern Analysis and Machine Intelligence 2007-12-20

The paper addresses the problem of recovering 3D non-rigid shape models from image sequences. For example, given a video recording talking person, we would like to estimate model lips and full face its internal modes variation. Many solutions that recover 2D sequences have been proposed; these so-called structure-from-motion techniques usually assume object is rigid. C. Tomasi T. Kanades' (1992) factorization technique based on rigid matrix, which produces tracking matrix rank 3 under...

10.1109/cvpr.2000.854941 article EN 2002-11-07

We approach the problem of stylistic motion synthesis by learning patterns from a highly varied set capture sequences. Each sequence may have distinct choreography, performed in sytle. Learning identifies common choreographic elements across sequences, different styles which each element is performed, and small number degrees freedom span many variations dataset. The learned model can synthesize novel data any interpolation or extrapolation styles. For example, it convert novice ballet...

10.1145/344779.344865 article EN 2000-01-01

This paper presents an inverse kinematics system based on a learned model of human poses. Given set constraints, our can produce the most likely pose satisfying those in real-time. Training different input data leads to styles IK. The is represented as probability distribution over space all possible means that IK generate any pose, but prefers poses are similar training data. We represent with novel called Scaled Gaussian Process Latent Variable Model. parameters automatically; no manual...

10.1145/1015706.1015755 article EN ACM Transactions on Graphics 2004-08-01

This article presents an algorithm for learning hatching styles from line drawings. An artist draws a single illustration of 3D object. Her strokes are analyzed to extract the following per-pixel properties: level (hatching, cross-hatching, or no strokes), stroke orientation, spacing, intensity, length, and thickness. A mapping is learned input geometric, contextual, shading features object these properties, using classification, regression, clustering techniques. Then, new can be generated...

10.1145/2077341.2077342 article EN ACM Transactions on Graphics 2012-01-01

We present a new method for creating an image with handpainted appearance from photograph, and approach to designing styles of illustration. “paint” series spline brush strokes. Brush strokes are chosen match colors in source image. A painting is built up layers, starting rough sketch drawn large brush. The painted over progressively smaller brushes, but only areas where the differs blurred Thus, visual emphasis corresponds roughly spatial energy demonstrate technique long, curved strokes,...

10.1145/280814.280951 article EN 1998-01-01

We present a new set of algorithms for line-art rendering smooth surfaces. introduce an efficient, deterministic algorithm finding silhouettes based on geometric duality, and segmenting the silhouette curves into parts with constant visibility. These methods can be used to find all in real time software. automatic method generating hatch marks order convey surface shape. demonstrate these drawing style inspired by A Topological Picturebook G. Francis.

10.1145/344779.345074 article EN 2000-01-01

Neural Style Transfer has shown very exciting results enabling new forms of image manipulation. Here we extend the existing method to introduce control over spatial location, colour information and across scale. We demonstrate how this enhances by allowing high-resolution controlled stylisation helps alleviate common failure cases such as applying ground textures sky regions. Furthermore, decomposing style into these perceptual factors enable combination from multiple sources generate new,...

10.1109/cvpr.2017.397 article EN 2017-07-01

This paper describes methods for recovering time-varying shape and motion of non-rigid 3D objects from uncalibrated 2D point tracks. For example, given a video recording talking person, we would like to estimate the face at each instant, learn model facial deformation. Time-varying is modeled as rigid transformation combined with Reconstruction ill-posed if arbitrary deformations are allowed, thus additional assumptions about required. We first suggest restricting shapes lie within...

10.1109/tpami.2007.70752 article EN IEEE Transactions on Pattern Analysis and Machine Intelligence 2008-03-27

This paper presents a data-driven approach to simultaneous segmentation and labeling of parts in 3D meshes. An objective function is formulated as Conditional Random Field model, with terms assessing the consistency faces labels, between labels neighboring faces. The learned from collection labeled training algorithm uses hundreds geometric contextual label features learns different types segmentations for tasks, without requiring manual parameter tuning. Our achieves significant improvement...

10.1145/1778765.1778839 article EN ACM Transactions on Graphics 2010-07-15

The style of an image plays a significant role in how it is viewed, but has received little attention computer vision research. We describe approach to predicting images, and perform thorough evaluation different features for these tasks. find that learned multi-layer network generally best -- even when trained with object class (not style) labels. Our large-scale learning methods results the published performance on existing dataset aesthetic ratings photographic annotations. present two...

10.5244/c.28.122 preprint EN 2014-01-01

This paper describes a simple technique to analyze Generative Adversarial Networks (GANs) and create interpretable controls for image synthesis, such as change of viewpoint, aging, lighting, time day. We identify important latent directions based on Principal Components Analysis (PCA) applied either in space or feature space. Then, we show that large number can be defined by layer-wise perturbation along the principal directions. Moreover, BigGAN controlled with inputs StyleGAN-like manner....

10.48550/arxiv.2004.02546 preprint EN other-oa arXiv (Cornell University) 2020-01-01

Camera shake during exposure leads to objectionable image blur and ruins many photographs. Conventional blind deconvolution methods typically assume frequency-domain constraints on images, or overly simplified parametric forms for the motion path camera shake. Real motions can follow convoluted paths, a spatial domain prior better maintain visually salient characteristics. We introduce method remove effects of from seriously blurred images. The assumes uniform over negligible in-plane...

10.1145/1179352.1141956 article EN 2006-01-01

This paper presents a novel physics-based representation of realistic character motion. The dynamical model incorporates several factors locomotion derived from the biomechanical literature, including relative preferences for using some muscles more than others. elastic mechanisms at joints due to mechanical properties tendons, ligaments, and muscles, variable stiffness depending on task. When used in spacetime optimization framework, parameters this define wide range styles natural human...

10.1145/1073204.1073314 article EN ACM Transactions on Graphics 2005-07-01

We introduce HuMoR: a 3D Human Motion Model for Robust Estimation of temporal pose and shape. Though substantial progress has been made in estimating human motion shape from dynamic observations, recovering plausible sequences the presence noise occlusions remains challenge. For this purpose, we propose an expressive generative model form conditional variational autoencoder, which learns distribution change at each step sequence. Furthermore, flexible optimization-based approach that...

10.1109/iccv48922.2021.01129 article EN 2021 IEEE/CVF International Conference on Computer Vision (ICCV) 2021-10-01

Abstract Portraiture is a major art form in both photography and painting. In most instances, artists seek to make the subject stand out from its surrounding, for instance, by making it brighter or sharper. digital world, similar effects can be achieved processing portrait image with photographic painterly filters that adapt semantics of image. While many successful user‐guided methods exist delineate subject, fully automatic techniques are lacking yield unsatisfactory results. Our paper...

10.1111/cgf.12814 article EN Computer Graphics Forum 2016-05-01

This paper presents a technique for computing the geometry of objects with general reflectance properties from images. For surfaces varying material properties, full segmentation into different types is also computed. It assumed that camera viewpoint fixed, but illumination varies over input sequence. one or more example similar materials and known are imaged under same conditions. Unlike most previous work in shape reconstruction, this can handle arbitrary spatially-varying BRDFs....

10.1109/tpami.2005.158 article EN IEEE Transactions on Pattern Analysis and Machine Intelligence 2005-06-22

We advocate the use of scaled Gaussian process latent variable models (SGPLVM) to learn prior 3D human pose for people tracking. The SGPLVM simultaneously optimizes a low-dimensional embedding high-dimensional data and density function that both gives higher probability points close training provides nonlinear probabilistic mapping from space full-dimensional space. is natural choice when only small amounts are available. demonstrate our approach with two distinct motions, golfing walking....

10.1109/iccv.2005.193 article EN 2005-01-01

This paper describes a photometric stereo method designed for surfaces with spatially-varying BRDFs, including both varying diffuse and specular properties. Our optimization-based builds on the observation that most objects are composed of small number fundamental materials by constraining each pixel to be representable combination at two such materials. approach recovers not only shape but also material BRDFs weight maps, yielding accurate rerenderings under novel lighting conditions wide...

10.1109/tpami.2009.102 article EN IEEE Transactions on Pattern Analysis and Machine Intelligence 2009-05-11

This paper introduces an approach to control of physics-based characters based on high-level features movement, such as center-of-mass, angular momentum, and end-effectors. Objective terms are used each feature, combined by a prioritization algorithm. We show how locomotion can be expressed in small number that balance is build controllers for human balancing, standing jump, walking. These provide numerous benefits: human-like qualities arm-swing, heel-off, hip-shoulder counter-rotation...

10.1145/1778765.1781157 article EN ACM Transactions on Graphics 2010-07-15

This tutorial describes several stroke-based rendering (SBR) algorithms. SBR is an automatic approach to creating nonphotorealistic imagery by placing discrete elements such as paint strokes or stipples.

10.1109/mcg.2003.1210867 article EN IEEE Computer Graphics and Applications 2003-07-01
Coming Soon ...