I am interested in computer vision and machine learning. I have been working on real-time visual detector and large-scale multimedia.
I am currently working on train simulation project at the Australian Center for Visual Technologies (ACVT), the University of Adelaide under the supervision of Professor Anton van den Hengel.
The ACVT conducts a challenging and difficult research in the area of image processing, computer vision and machine learning.
Our objective is to design a new technology that can match the full capability of human vision.
A list of publications.
Optimizing the partial area under the ROC curve
a novel ensemble learning method which achieves a maximal detection rate at a user-defined range of false positive
rates by directly optimizing the partial AUC using structured learning.
S. Paisitkriangkrai, C. Shen and A. van den Hengel. Efficient Pedestrian Detection by Directly Optimizing the Partial Area under the
ROC Curve, ICCV 2013, PDF Supplementary
Greedy Sparse Linear Discriminant Analysis (GSLDA) for its
conceptual simplicity and computational efficiency to the
task of face and pedestrian detection.
S. Paisitkriangkrai, C. Shen and J. Zhang. Sharing features in multi-class boosting via group sparsity, CVPR 2009, PDF Journal
Multi-class Boosting + Group sparsity = Feature sharing classifier
We present a novel formulation of fully corrective boosting for
multi-class classification problems
with the awareness of sharing features.
S. Paisitkriangkrai, C. Shen and A. van den Hengel. Sharing features in multi-class boosting via group sparsity, CVPR 2012, PDF Supplementary
|AdaBoost.ECC||Our MultiBoost (group)|
Multi-class boosting with Crammer and Singer's formulation
We propose scalable and effective classification model for multi-class boosting classification.
The approach directly maximizes the multi-class margin (similar to the multi-class SVM (MSVM) formulation of Crammer and Singer).
The technical paper and source code for MultiBoost (Stage-wise) can be downloaded here. PDF source code.
|MultiBoost (Stage-wise)|| |
Randomized boosting with matlab source code
The new model trains a single-vector parameterized classifier irrespective of the number of classes.
The advantage of the new appraoch is that multi-class boosting can be trained at the same learning complexity of binary boosting.
The technical paper and source code for RandomBoost can be downloaded here. PDF source code.
Large-scale multimedia framework
Binary-class visual detector
Machine learning demo
Voronoi Diagram / Delaunay Triangulation
Weka download (java -Xmx1000M -jar weka.jar)