Ashley Eden
545 Soda Hall
University of California, Berkeley
Berkeley, CA 94720-1776
Email: eden 'at' cs.berkeley.edu
Ashley Eden 545 Soda Hall University of California, Berkeley Berkeley, CA 94720-1776 Email: eden 'at' cs.berkeley.edu |
|
 |
A
Method for Cartoon-Style Rendering of Liquid Animations Ashley Eden, Adam Bargteil, Tolga Goktekin, Sarah Beth Eisinger, James O'Brien To appear in Graphics Interface, 2007 We present a method for
cartoon-style rendering that demonstrates how the output of a liquid
simulator can be used to drive a compelling cartoon-style liquid
animation. Our method is based on four cues that emphasize
properties of the liquid surface's shape and motion. It is fast,
easily tuned, and portable to a variety of mainstream liquid simulation
systems and renderers.
|
 |
Seamless Image Stitching of Scenes with Large
Motions and Exposure Differences Ashley Eden, Matthew Uyttendaele, Richard Szeliski Computer Vision and Pattern Recognition (CVPR), 2006 We present a technique to create a
high dynamic range (HDR) panorama given images with large exposure
differences and large motions within the scene. We also require
no extra hardware attachments to the camera. We introduce a
two-step graph cut approach: the first step fixes the position of
moving objects within the scene, and the second step fills in the
dynamic range.
|
 |
Video
Based Motion Synthesis by Splicing and Morphing
Greg Mori, Alex Berg, Alyosha Efros, Ashley Eden, Jitendra Malik Report No. UCB/CSD-4-1337 I helped out on this research, which was a method of synthesizing videos by splicing together clips of input videos. It first introduces "kinematically correct morphing", which allows for input clips to be smoothly spliced together. It also contributes novel activity recognition algorithms, used to automatically label the input data so that the synthesized video can be controlled with high-level labels. We demonstrate the technique on ballet and tennis clips. |
 |
This is a class project in which we applied computer vision methods to the task of automatically predicting human attractiveness from frontal face images. A dataset of thousands of images and corresponding attractiveness scores was obtained from a popular website. Using a combination of radial basis functions and specialized feature detectors, we achieved moderate success in predicting female attractiveness. |
 |
Eliminating
Ghosting and Exposure Artifacts in Image Mosaics
Matthew Uyttendaele, Ashley Eden, Richard Szeliski Computer Vision and Pattern Recognition (CVPR), 2001 We introduced a method for automatically stitching together image mosaics in the presence of moving objects and exposure differences. If moving objects between overlapping images are left in, they will appear "ghosted". Treating such regions as nodes in a graph and applying vertex cover allows us to remove all but one instance of each object. We also continually adjust exposure across images in order to eliminate visible shifts in brightness or hue. We calculate exposure corrections on a block-by-block basis, and smoothly interpolate the parameters. |
 |
This is my undergraduate thesis. I present an algorithm for motion synthesis based on a popular texture synthesis technique. Essentially, we can synthesize a new "random" motion in the general style of the original motion, or with pose and/or specific style constraints. Given the character's (in this case a dog's) hierarchical skeleton and bone lengths, and input motion capture information, we generate realistic motions similar, but not limited to, the content of the sample motions. |