First Time To Teach

This past thursday (12/10/2009) I had to opportunity to teach the Advanced Vision class because my advisor was out of town and couldn't teach the class. It was a really great opportunity to see what it's like to lecture for the entire class compared to just giving a presentation.

I ended up being a little short on time, took a bit of the rest of it to answer questions from the class and go into further detail about my research.

Attached in the presentation I gave.

MotionTracking ACV2009 Winter

Invisible

Very interesting talk about the invisible.

Used as an example

I had a neat experience recently (this past quarter). My old professor from when I took Artificial Intelligence asked if he could use my paper as an example for the class. That's the first time a professor has asked to use my work as an example for others.

Here's the link. Look for "Example Research Paper".

Computer Vision Paper

Here's a direct to the paper (pdf)

PDF Paper Link

Computer Vision Object Tracking Spring 2009

Publish at Scribd or explore others: object tracking computer vision

Computer Vision

Well, it's been quite a while since I've had a post.

So, what have I been up to? I guess a video would be more helpful than what I could say.

I've been working on Computer Vision. Specifically I've been working on object tracking. I did it as a small project for an Introduction to Computer Vision class and continued it with an Advanced Computer Vision class and an Independent Study. It's been quite a lot of work, and learned a lot along the way. Also, here's a YouTube playlist of some of the other results that I played around with, some better than other.

YouTube Playlist

I have a research paper as well documenting my implementation and results that I'll post soon!

Any Questions? leave a comment

First Research Paper

Hi all, Long time no content. Anyway, I've been fighting through my first quarter in grad school and all I have to show for it is a crummy research paper. Anyway, if you're interested, here's a link to it:

http://www.scribd.com/doc/8096612/CBIR-Research-Paper?secret_password=22uh3l6h29p36v34eg5l

It's on CBIR (Content Based Image Retrieval) or being able to search images based on image content, not needing image tags or other meta data.

Enjoy

Reducing the search space

For a class I'm taking, we're looking at Constraint Satisfaction Problems or CSPs. For a homework assignment we're looking at different methods of solving these CSPs. One way to do this is to create a tree of the possible variables in the CSP and at each level assign values. One obvious optimization is to reduce the assigned values at each level to only those that meet the constraints. According to the text this reduces the search space from n!*d^n to d^n where d is the size of the domain and n is the number of variables in the CSP.

As much as I'm interested in theory, practice is very helpful in explanation. One problem is the following:

  TWO
+ TWO
------
  FOUR

This is a crypt-arithmetic puzzle where a unique value is assigned to each letter, and no word can have leading 0's. To see this for myself I quickly implemented this in ruby to see the difference between constraining the search at each level and not. Results?

Unconstrained search (should be n!*d^n):

real 0m44.690s
user 0m42.769s
sys  0m0.315s

Constrained Search (should be reduced to d^n)

real 0m0.049s
user 0m0.037s
sys  0m0.005s

As you can see, a huge difference. Even though I fully understood the math, I was surprised how much of a difference it made by reducing the search space (even though it involved more checks at each level).

In summary... reduce