TITLE:

From 2D Image Recognition to 3D Scene Reconstruction: A Restricted Turing Machine

DATE:

Friday, October 16, 2015

TIME:

3:30 PM

LOCATION:

GMCS 214

SPEAKER:

Dr. Xiaobai Liu. Assistant Professor of Computer Science, San Diego State University

ABSTRACT:

Computer Vision aims to answer two types of questions – what and where – that involve 2D

Image Recognition and 3D Scene Reconstruction, respectively. This talk will cover several

studies for solving these two tasks, either separately or jointly. It includes four major parts.

Firstly, I will present an attributed grammar model for simultaneous 2D recognition and 3D

reconstruction from single view, which achieved state-of-the-art scene reconstruction

results on several public benchmarks. Next, I will introduce how this unified framework

can be applied to solve a number of novel problems, most of them are based on moving

camera platforms (e.g. Robots). Thirdly, I will demonstrate a vision-based intelligent

system, i.e. Turing Machine, which is capable of understanding large-scale videos, like

human beings, and answering written questions. Last, I will discuss a few research

directions that are potentially valuable in both academy and industry.

HOST:

Dr. Jose Castillo

DOWNLOAD: