Sure i`ll let you know once we confirm the findings ... still at the initial stage but pretty exciting if its correct :)

Here I have a slice of my data with the y (forward) direction being to the right and the time (consecutive frames) being downwards. Now I want to use these contours to determine to which cluster a contour belongs

Any suggestions on how to get the min/max values of y for every frame?

I guess I could just ask OpenCV to give me a polygon that has every coordinate

There we go

I added some erosion, to make sure the two blobs are separated, so it doesn't fit exactly, but now I can assign each contour in the x,y plane to one of the contours in the y,z plane

Closer, getting closer

@IvoFlipse @Kortuk Hi all

hi @Phonon

@Phonon allo mate!

slowly making progress

@IvoFlipse Nice. I see pretty pictures.

Yeah, I've been plotting my data in a different plane and now I'm trying to map the contours I find in the x,y plane onto the contours I find in the y,z plane

Because as you can see those contours are (linearly) separable, I just need a way to abuse that fact

@IvoFlipse neat.

But I haven't managed to get it working properly yet :P

@IvoFlipse I submitted the Algorithms assignment 40 minutes before deadline last night = )

@Phonon What was the exercise this week?

@IvoFlipse Strongly connected components

Ah yes, Kasaraju right? That one was brutal

@IvoFlipse Yeah. I actually had it working, but then I started getting stack overflows on the actual input.

I got bitten by Python's way of dealing with pointers and references (or rather not)

@IvoFlipse Solution: reboot into Linux, which has a 16MB default stack as opposed to 1MB stack in Windows. = )

@Phonon Yup, you could have read about my problems with the exercise on chat :P

I added some recursion depth, but mostly I had to stop copying endlessly I believe

Ah yes, my problem was that I tried to pass along the list of items I already explored, but I kept overwriting it, so my bookkeeping went to hell

I ended up making them globals, which makes me feel dirty but did the trick

@IvoFlipse I'm surprised how quickly my algorithm ran though. I had an old Linux machine running a Dual Core Pentium processor. Did it in 10 seconds.

And I had to do some nasty stuff with calling threading

Yeah its actually not so complicated, but you have to get the details right

@IvoFlipse Yeah, I'm glad I didn't have to resort to that.

Hmm it seems my Python implementation is a lot slower than yours :P

ah well, I felt great after I managed to solve it :D

@IvoFlipse Well, Python is scripting. I did mine in C.

That explains a lot

@IvoFlipse Yeah, I was really hyped when my answer got accepted =)

I slightly regret not trying it in C++, because that way at least I could have managed what variables it was abusing

@IvoFlipse I was thinking of C++ as well, but I'm forcing myself to learn good C practices, so I stuck with it. = )

@Mohammad Sup dude! I'm working on that email! = )

Hehe, well I could use the practice

@Phonon Hey dude! HEhe, ok ;-)

@IvoFlipse @GauravJain Hey all

hi @Mohammad

@Kortuk ola

@Phonon No worries, I had planned to do some DSP reading over the weekend... but got distracted by the fairer sex... :P

@Mohammad allo

@Mohammad Best kind of distraction IMO. Other than DSP itself of course. \Dork Alert\

@Phonon Well they can be very distracting, especially when you just moved. Suddenly I have to help hang up paintings and move plants around. Or drive to Ikea to get yet another closet

@Phonon lol yeah. "Uhh..sorry babe I cant Im reading umm... about IIR filters..." 0_0

@Mohammad Happened to me before.

@Phonon I have similar stories... If there was any doubt about the nerdiness inherent here, today it was killed. lol

@Phonon @IvoFlipse So remember how I said that ICA uses knowledge of a source signals' PDF? Well, apparently you can use the sigmoid function to model most real life signals' CDFs...

Who knew! lol

@Phonon @IvoFlipse @Kortuk I really wish there are papers/books describing how people come up with algorithms and the like - its not enough for us to know how they work, but we also need to know how exactly they decided/stumbled upon their insights. That would go a long way

@Mohammad I didn't :P

@IvoFlipse Have you used elance.com by any chance?

I have not

@Mohammad Usually finding the original papers helps a bit. Original paper on ICA is pretty good.

@Phonon It helps if you actually have access to papers :(

Almost makes me want to become a member of ACM

@Mohammad @IvoFlipse Pretty sure that that's the one. At least it's by the same authors.

bsp.brain.riken.jp/ICApub/NN00.pdf

@IvoFlipse ACM?

acm.org

@IvoFlipse I don't think the the membership gives you full database access. At least that's how IEEE works.

@Phonon Yeah thats true. Although I have noticed a lot of times authors will say, "We seek to do this". "It turns out this formula is a good way to measure this quantity". Well, how did they stumnble on that ?

@Phonon Link appears broken btw...

@Mohammad Hours upon hours upon hours upon hours upon hours of PAIN usually. = )

I guess so @Phonon, but at least their library is pretty relevant for me

redwood.psych.cornell.edu/discussion/papers/…

try this

@Phonon hehe

Darn I wish I was better at poking around, I always like to check what articles they have available on such pages :P

redwood.psych.cornell.edu

The base url seems pretty comprehensive too

@Phonon Oh I already read that paper lol

@Phonon Actually I ordered his original book last week too

@Mohammad Nice. Man, at this pace I'll need to get yet another bookshelf soon.

@Phonon I buy books like a woman buys shoes... and guilty of reading a little and not touching them for 6 months after that too...again just like women with her shoes..

@IvoFlipse Hmm yeah, looks interesting...aaannnddddddddd bookmarked.

@Phonon Hehe, you know I just noticed how much of a beat up Wiener filtering gets - almost every paper I see about denoising has Weiner filtering image as a 'oh looks how sucky this looks compared to our denoising". hehe

@IvoFlipse @Phonon Interesting paper from ivos link (redwood.psych.cornell.edu/papers/olshausen_field_2000_amsci.pdf)

Interesting indeed

@Mohammad Yeah, same here. Even though I started reading more lately. So I don't feel as bad.

How to separate these two :\

What are you trying to separate?

What you see are two paws landing very close after one another, so close even, that a part of the hind paw is making contact within the area where the front paw made contact before, there's perhaps only 1 pixel of space between them

Note that the red area you see has been eroded to create a clearer gap

But the algorithm is wrongly merging them, because it thinks they are so close together, they must be part of the same paw

I tried using this information to track the contours directly, but that went horribly bad, so now I'm back at my previous attempt, but I want to post process the paws I found in the first pass and separate them based on the side way view

@Phonon @IvoFlipse Thats it. Im doing a PhD in neuroscience+dsp

@Mohammad lol, good idea

@IvoFlipse @Phonon I just realized it...thats what Ive been after all along... not learning some new algorithm for its own sake... no...but for the linking of our algorithms to what possible algorithms nature herself has already figured out... that link - that relationship..is what I am truly after..

@Mohammad Good approach. Although some algorithms, transforms and so on are not necessarily nature-inspired. Wavelets are considered to be a quite artificial transform. Just as an example. But yeah. My approach is that if my brain can do something, then there's enough information for a computer to do it as well.

@Phonon Certainly, certainly. They might not overlap completely. However, just reading that article I linked, the retina and some support neurons at V1, already hit on Fourier, Wavelets, sparse coding, image proc (histogram equalization) (!) - its truly fascinating to see that link with what evolution has come up with. Intersting to explain nature using our vocabulary, and if we cannot, it can teach use new vocabulary as well

*us

@Mohammad Yeah, definitely.

@Phonon @IvoFlipse coursera.org/course/maththink

@Mohammad Muahahahaha! I've already done combinatorics, real analysis and number theory (finishing right now). Challenge accepted! = )

@Mohammad Signed up!

@Mohammad @IvoFlipse Algorithms II is up! October! Yay! = )

First to Like of Facebook! = )

Meh... I think I had one too many of those free cookies for lunch. Coursea excitement + sugar rush = me totally hyper.

@Phonon You did all those?! Ok Im officially jelly.

@Phonon And its decaf coffee for me here since after 12pm!

@Phonon Are you doing that stats class btw??

@Mohammad I totally with you on decaf

@Mohammad Thinking about it.

@Mohammad I'm signed up though

@Mohammad there are no programming assignments for stats, so may as well do it,.

@Phonon Hmm interesting

@Phonon What are the hws then?...

@Mohammad Just practice questions I guess. But there is no mention of programming on the web site.

@Phonon aaannnddddd enrolled. (stats from udacity)

@Mohammad Ah, I was talking about Coursera one. Sorry = )

@Mohammad Starts in September

@Phonon Ah!

@Phonon Dude I wish there was a job where like, they pay you to learn lol

@Mohammad I'm afraid it's the other way around. You learn to be able to get paid. = )

@Phonon I know, although I am finding out even the big bucks arent satiating my curiosity! :-)

@Mohammad Its a tad disappointing, I had hoped it would be more like CS101

@IvoFlipse I still haven't even started it lol.

Testing (luckily) has a lot of talking, not continuous quizzes

@IvoFlipse Which one?

But its a bit soft anyway, though the homework can be quite annoying, trying to get full coverage om some piece of code you don't know

Udacity's CS258

Anyway, dinner time, bbl

@Phonon What class is that?

@Mohammad Udacity stats. I'm signed up, but I haven't even visited the page yet. Busy with algorithms and number theory.

Downloading Jelly Bean update for my phone. Yay! = ))) I've been compulsively pressing the "Check updates" button every two hours for the past three days.

@Phonon What is that??

@Mohammad Latest Android OS.

@Phonon Hehe I already installed it right after Google I/O

strange coursera doesn't seem to work on Chrome anymore :S

@IvoFlipse really?

they won't finish loading for me

@IvoFlipse That sucks. Must be new. I run Chrome at home.

I'm using the dev version though

hmm

Interesting

classx.stanford.edu/ClassX/system/users/web/pg/…

Funny the Dutch accent by the lecturer: explorecourses.stanford.edu/instructor?sunet=gerritsn

Probability for CS students classx.stanford.edu/ClassX/system/users/web/pg/…

and I should one day follow that Convex OPtimization course

@Phonon @IvoFlipse OMG!

@IvoFlipse where do you see that convex opt class?