Serge Belongie
06062010
02:52 AM PT (US)

/m32 The final exam will take place 36pm on Tuesday June 8 in PETER 103 (the usual classroom).

Justin
06022010
11:21 PM PT (US)

Where will the final exam be held, again? Thanks.

Eric Christiansen
05292010
03:45 PM PT (US)

/m30 Hey Chris, Serge says he prefers handwritten, but notes can be typed as long as you type your own notes (you can't just shrink the scribe notes down to 2 pages). In that case, the minimum font size is 9pt.
BTW, 2 pages is 2 pages front and back. So if you print doublesided, that's 4 8.5x11 surfaces.
Also, it's okay to collaborate when making note sheets. Just remember that Serge wants each individual to write up her note sheet on her own.

Chris
05292010
02:44 PM PT (US)

I have a question about the final. Looking at previous finals they say we can bring a calculator and two 8.5” by 11” sheets of notes to the exam. Can these notes be typed? Thanks.


Deleted by author 05132010 05:18 PM

Prasanna
05132010
12:10 PM PT (US)

Are you talking about the "Simple Calibration device" problem?

Eric Christiansen
05132010
11:57 AM PT (US)

/m25 When I did this problem, I experienced some instability with respect to which pairs of orthogonal lines I chose. You might want to try choosing the same lines that were chosen in Figure 2.17 on page 57 of H&Z.

Prasanna
05132010
10:34 AM PT (US)

/m25 Yeah I set the complex numbers to real and the output is a singular matrix. I suspect it is the initial H(still wondering the possibility of H being complex valued's meaningbecause for other images my function gives real values) that is causing the problems. But as a way of check I backward calculated the coefficient matrix of the image of the absolute conic. They were:0.0000 0.0000 0.0004 0.0000 0.0000 0.0003 0.0004 0.0003 1.3582 as compared to my values of: 0.0000 0.0000  0.0000i 0.0001  0.0000i 0.0000  0.0000i 0.0000  0.0000i 0.0002  0.0000i 0.0001  0.0000i 0.0002  0.0000i 1.0000 + 0.0000i

Eric Christiansen
05132010
08:58 AM PT (US)

/m24 I don't think you should be getting imaginary values. Have you tried simply throwing the imaginary parts away? If the imaginary parts are small, then you'll still end up with an approximate null vector.

Prasanna
05132010
06:35 AM PT (US)

I have doubt in the q.4. I obtain a H that has imaginary values. Is that allowed? The values in the KK^{T} is close to the value from KK^{T} given in the text. But it has small imaginary values that are on some of the elements. As a result, cholesky gives an error. Do you have guess where I may be going wrong?

Eric Christiansen
05122010
04:54 PM PT (US)

/m21 Yes, you should assume the ^{T} refers to the conjugate transpose.

Sayanan
05112010
05:55 PM PT (US)

This may have been obvious to others in the class, but on my machine at least, using Hartley Normalization for fitting the conic in problem 4 made a huge difference.

Sayanan
05102010
07:54 PM PT (US)

In HW3, #2, we use the (*) notation to denote the conjugate transpose of a vector or matrix. We use the ^{T} notation to denote the transpose.
My question is the following: Should we assume that the ^{T} extends to a ^{H}, Hermitian transpose, considering that we're working with potentially complex matrices here?
I believe that this has a bearing on part (b)'s derivation, as far as grouping/distributing the Hermitian operator goes.

Eric Christiansen
05052010
10:15 PM PT (US)

/m19 If what the function does is clear from context or its name, and you submitted it in a previous assignment, you don't need to submit again.

Chris Kanan
05052010
09:42 PM PT (US)

For homework 3, should we include functions written in previous homeworks in an appendix? Right now I'm just putting comments in the new code denoting that a function was written for an earlier assignment.

Eric Christiansen
04292010
12:40 AM PT (US)

/m17 I think you can avoid cleverness in the form of cross and dot products by using lsqlin, an iterative least squares solver from Matlab's optimization toolbox.
