Twitter Influential Graph

On social networks, people interact with other people in many different ways. By analyzing these interactions, we can estimate how a user is influenced by other users on a social network. For example, a Twitter user may respond—retweet, like or reply—tweets that are published by other people. The tweets that he/she responded will appear on his/her public profile page, with the information of the user who published the tweets. Using this information, we are able to determine the influential level of a user to another user.

Semantic Similar Image Retrieval

The system tries to solve the problem that the intention of the end user is unknown by a content-base image retrieval system. The system implement a full chain of image retrieval system, include feature extraction, quantization, hashing, retrieval, graph search, etc. The implementation uses Python, Matlab, SQL, HTML/CSS, Javascript, and bash scripts.

Human Pose Estimation

This system uses a deep learning algorithm to estimate the position of the joints on a human body from images.

People Segmentation and Tracking

This system segment people body and track the body movement. The system is based on faster-rcnn model.

Actors/Actresses Recognition in Movies

This system uses deep neural network and pre-trained face model to recognize actors/actresses in movies. The system only detect and recognize frontal faces. About 1M celebrities can be recognized.

Face Anti-spoofing

Face spoofing attack is a situation that an unauthorized party uses face images or videos of an authorized user to illegally gain access to a system. For example, one can use a face image on another mobile phone to unlock Samsung Galaxy S8.

In this demo, we show our research result on face anti-spoofing. Our system is based on deep learning framework. In this video, genuine faces are shown with green box; spoofing faces are shown in blue box.