Research

Broadly speaking, I am interested in computer vision, machine learning, applied deep learning and multimedia related research problems as below.

- Image understanding (image parsing, image re-targeting, image re-attentionizing)

- Visual fixation analysis with image, video, audio, depth, touch behavior

- Salient object detection

- Action/gesture recognition

- Human biometrics (human attractiveness, human weight prediction)

- Augmented reality and virtual reality

- Deep learning

I like doing something cool. Please check some demo videos below.

RESEARCH INTERESTS

- Computer Vision

Camouflaged Object Segmentation: [Visit Project Page]
Salient Object Detection with Semantic Priors: [Visit Project Page]
Salient Object Detection with Augmented Hypotheses: [Visit Project Page]
1. Twin Recognition: [Visit Project Page]
Beauty Sense - M2B dataset [Visit Project Page]
NUS3D-Saliency Dataset [Visit Download Page]: We collect a large human eye fixation database compiled from a pool of 600 2D-vs-3D image pairs viewed by 80 subjects, where the depth information is directly provided by the Kinect camera and the eye tracking data are captured in both 2D and 3D free-viewing experiments. This work was published in ECCV 2012.
Discovery.news "The Future of Intelligent Ads: Reading Your Weight" reported my work "Seeing Your Weight – An application in targeted advertisement" shown in ICCV'11 as a demo. A detailed paper is published at Journal of Computer Science and Technology [Visit Project Page].
Action Recognition [Visit Project Page] We propose a new pooling method for action recognition based on visual saliency. This work is published in IEEE Transactions on Circuits and Systems for Video Technology.
Scene Parsing. We propose a new image parsing method based on global and local CNNs. The test image is first oversegmented into superpixels. Then the similar images are retrieved by the global-level CNN feature matching, and the class label of each superpixel is initiated by superpixel classification which involves regional-level CNN and hand-crafted (HC) features. The initial labels, in combination with exemplar detection results, followed by contextual smoothing, provide dense labels of each pixel in the test image. This work is published in ICARCV 2016.