Dr. Jun Liu PhD in Computer Science Office 2.90, Kilburn Building University of Manchester, Oxford Road Manchester, M13 9PL, UK Email: jun@cs.man.ac.uk As of November 2007, I passed my PhD thesis viva and will join The Foundry, working on visual effects and image processing technologies for film post-production. From April to October 2008, I will also work in the Sigmedia group, Trinity College Dublin as a Marie Curie Postdoctoral Research Fellow. This web page will therefore be outdated while I'm having fun there. Please contact me at junxliu@gmail.com.

Short Bio

Jun Liu joined the Department of Computer Science, University of Manchester as a PhD student in September, 2004. He is a member of the Advanced Interfaces Group working with Professor Roger Hubbold. His main research interest is in general areas of computer vision and graphics and more specifically in "automatic 3D model generation from multiple 2D images". His research is funded by the departmental scholarship.

Before coming to Manchester, Jun Liu studied at Zhejiang University, PR China where he was awarded his Bachelor's Degree (with Honours) in Computer Science in June, 2004. From May 2003 to June 2004, he worked as an undergraduate research assistant at Centre for Engineering and Scientific Computation with Professor Yao Zheng, mainly on tiled stereo display system.

Publications

A Perceptually Validated Model for Surface Depth Hallucination M. Glencross, G. Ward, C. Jay, J. Liu, F. Melendez and R. Hubbold ACM SIGGRAPH 2008 PDF (1.6M) Project Page (with demo videos)

Mesh Optimisation Using Edge Information In Feature-Based Surface Reconstruction Jun Liu and Roger Hubbold International Symposium on Visual Computing (ISVC'06) PDF (5.96M) | BibTex A new method to refine and optimise surface meshes using edge information in 2D images A meshing -- edge point detection -- remeshing scheme that can gradually refine the surface mesh until it best fits the true physical surface of the object being modelled.

Automatic Camera Calibration and Scene Reconstruction with Scale-invariant Features Jun Liu and Roger Hubbold International Symposium on Visual Computing (ISVC'06) PDF (5.29M) | AVI (38M, divx) | BibTex Effcient high-dimensional nearest neighbour search Framework for structure from motion using SIFT Solved "phantom point" problem and greatly reduced the possibility of error propagation

Activities

I am a member of IEEE Computer Society.

I am a committee member of SIGGRAPH Professional Chapter in Manchester.

I was a paper referee for the following conferences:

• 5th IEEE and ACM International Symposium on Mixed and Augmented Reality (ISMAR'06)
• 17th British Machine Vision Conference (BMVC'06)

I was a paper referee for the following journals:

• IEE Proceedings - Vision, Image, and Signal Processing

I attended various computer science conferences:

• 16th British Machine Vision Conference (BMVC'05),
  5-8 September 2005, Oxford, UK.
• 10th International Conference on Computer Vision (ICCV'05),
  15-21 October 2005, Beijing, China
• 2nd European Conference on Visual Media Production (CVMP'05),
  30 November-1 December 2005, London, UK
  (I was awarded the EPSRC Student Bursaries)
• 2nd International Symposium on Visual Computing (ISVC'06),
  6-8 November, 2006, Nevada/California, USA

I am a member of the PhD Mentors Team in the department.

I was also involved in teaching of the following subjects:

• CS1580 Object Oriented Programming with Java
• CS2341 Software Engineering
• CS2072 Computer Graphics
• CS644   Advanced Machine Vision (MSc course)

Downloads

I've been developing software almost exclusively in C++ under Linux, although occasionally in Matlab (most often Symbolic Toolbox) for algorithm validation. The software libraries I frequently use include: VXL, Boost, OpenGL, and QT4.

The code below is absolutely FREE (as both in free speech and in free beer). The code is written in the hope that it might be useful to the research community, but without warranty of any kind.

If you find the code useful, please let me know by sending me an email, whether it is a description of its usage in your project, bug report, or simply a piece of joke you find funny.

C++ source code of SIFT	C++ implementation of scale-invariant feature transformation (SIFT) tar.gz (46K) | zip (48K)
SfM test case generator	Generate test cases for structure from motion, using 3D point cloud from PLY file tar.gz (1.5M) | zip (1.5M)
Visualisation programme	Visualise and navigate the points and cameras tar.gz (17K) | zip (23K)

Demos

Automatic video rotoscoping

Non-photorealistic rendering on existing videos to obtain cartoon-like effect. The process runs 3 FPS on a 2.0 GHz CPU. Examples include film clips from The Matrix and Amelie.

Miscellaneous

Photo album

Conference deadline calendar

---

Last updated: 22nd December, 2007