Computer Graphics Laboratory ETH Zurich

ETH
CGL » Research » Digital Humans

Digital Humans

Introduction

Mastering and understanding human faces generally encompasses a variety of different research challenges. We investigate novel approaches to acquire and represent human faces, and to develop algorithms that enable stunning visual effects and even medical applications.

Topics

Geometry Acquisition & Reconstruction

Capturing and reconstructing the shape of the human face with very high spatial resolution and fidelity is an essential part of our research. We are currently focusing on purely passive systems. Our current system is a scalable single-shot scanner that consists of off-the-shelf consumer cameras.


Eye Capture

The human eye is one of the central features of individual appearance. To faithfully reproduce all its intricacies we propose a novel capture system that is capable of accurately reconstructing all the visible parts of the eye: the white sclera, the transparent cornea and the non-rigidly deforming colored iris.


Representations and Models

The demand for light weight facial performance capture requires developing new ways to represent and model the human face. Our anatomical-constrained face model can capture extreme deformation of the face from smartphone camera input.


Performance Capture and Tracking

The shape of the human face changes due to speech and facial expressions. Capturing and tracking the dynamic shape of the face in very high detail is challenging, due to the rapid and drastic changes the face undergoes. We developed a method to capture a dynamic performance with high fidelity using video cameras.


Synthesis and Augmentation

The synthesis of realistic human faces prevails throughout the entertainment industry, from virtual characters in films to real-world animatronics. Many motion capture and animation techniques can only produce low-resolution animation, lacking dynamic fine-scale detail. We developed techniques to enrich these low resolution sequences using an expressive model captured from and for each actors.


Publications





© CGL, ETH Zurich