|
Yi Li Researcher, Computer Vision Discipline, NICTA L1-31, Tower A, 7 London Circuit, Canberra, ACT 2601 firstname dot lastname AT cecs DOT anu DOT edu DOT au or firstname dot lastname at nicta dot com dot au |
|
|
· Bio |
|
|
|
Education · 2004 – 2010 Computer Vision Lab, University of Maryland, College Park Advised by Prof. Yiannis Aloimonos and Dr. Cornelia Fermuller · Computer Vision and Machine Learning · Visual Perception and Optical Illusions; · Action and Object Recognition; · Sparseness Recovery;
· Jan
2008, Future Faculty
Fellow, A. James Clark · 2nd place, 1st Semantic Robot Vision Challenge (sponsored by NSF), AAAI2007, Vancouver, Canada, 2007. |
|
· Courses in NICTA |
|
|
How to do Computer Vision research |
Seminar Room D, Level 3, Wednesday 3pm, May 4-May11 Click here for the evaluation form. |
|
Selected topics in Computer Vision |
Seminar Room D, Level 3, Wednesday 3pm, May 18-July 6 Link to the syllabus |
|
· Students I am actively seeking
motivated graduate students and visiting students around the world. NICTA is able
to provide a full or top-up scholarship for PhD students and support visitors
in various forms. Please send me your resume if you are interested in
collaborating with me (though your formal application must through ANU
admission office if you plan to pursue a graduate degree). Chinese Scholarship Council (CSC) students: Australian National
University and the China Scholarship Council (CSC) have a collaborative
arrangement to provide research opportunities to select high quality research
students from China. CSC scholarship recipients will be able to enrol in the
ANU PhD program for up to four years full-time study. NICTA also welcomes the
CSC visitors to collaborate in many projects. |
|
|
PhD (ANU) |
Kyoungup Park (with Nick Barnes) Song Wang (with Nick Barnes) |
|
Master (honour) |
Ashley Stacey |
|
Bachelor (honour) |
Tianxing Li |
|
Visiting (CSC) |
Fang Wang (NUST) |
|
· Research |
|
|
|
A.1. Modelling Where Humans Look in a Picture We proposed a supervised method for modeling human’s attention in a picture without using any high level detectors. This machine learning module extends the supervised Latent Dirichlet Allocation (sLDA) by augmenting image features as additional side information for cue integration. Each pixel in an image is regarded as a “document” and all possible feature histograms as “vocabulary” in the topic modeling. The main idea is to learn the topics of documents (pixels) and further infer the probability of saliency. Experiments showed that this method is competitive with the state of the art algorithm that used three effective object detectors on the largest available dataset. Further experiment on cross-dataset validation suggested that our approach models successfully where humans focus in a picture. |
|
|
A.2. Integrating
Eye Trackers and Head Mounted Display for Navigation Experiments.
Navigation using salient information is an important component in the Bionic Eye because the current hardware simulation can only generate low resolution images. To study how humans navigate using band limited information, we integrate an eye tracker to a head mounted display in the navigation task. |
