Die Suche ergab 25 Treffer

von diane
21. Jul 2010 17:44
Forum: Computer Vision
Thema: Probabilistic Recognition
Antworten: 4
Zugriffe: 1046

Re: Probabilistic Recognition

A local measurement M_k (not necessary a patch, but a value extracted in a given location) is extracted from an image. p(M_i|O_n) is coming from learning. It is the normalized histogram which represents the probability density function of object O_n. There is no distance measure involved in the trai...
von diane
20. Jul 2010 21:56
Forum: Computer Vision
Thema: Probabilistic Recognition
Antworten: 4
Zugriffe: 1046

Probabilistic Recognition

I would like to make some details clearer here for the people asking a question this afternoon. During the office hour today, what we discussed about is the general principle of global histogram based probabilistic recognition. In the slides "histograms", from slide 39 to 52, a more sophisticated fr...
von diane
20. Jul 2010 19:39
Forum: Computer Vision
Thema: About the Harris determinant
Antworten: 0
Zugriffe: 386

About the Harris determinant

A question I had today: Why is the determinant of the matrix not 0 in the Harris detector? If you look at the interest points slides, slide 17, you can see that you need to apply the Gaussian convolution to the 2x2 matrix before extracting the determinant. The illustrations on the same slide already...
von diane
20. Jul 2010 13:15
Forum: Computer Vision
Thema: Error in Solution Ex1 houghtransform.m?
Antworten: 2
Zugriffe: 455

Re: Error in Solution Ex1 houghtransform.m?

About the first question: You are right, the current proposed solution is not fully satisfactory as it does not distribute the values between -D and D equally, ie some bins are more populated than others. But it does use the correct range of values. The solution you mention, ie replacing round by fl...
von diane
20. Jul 2010 09:48
Forum: Computer Vision
Thema: Error in Solution EX2 normalites_hist.m?
Antworten: 1
Zugriffe: 498

Re: Error in Solution EX2 normalites_hist.m?

You are right, the white color needs to get a special treatment with this solution.
You can also replace 255 by 256 and the formula holds.
von diane
15. Jun 2010 19:21
Forum: Computer Vision
Thema: Project report - questions
Antworten: 1
Zugriffe: 544

Re: Project report - questions

1. I don't get fully the question. Of course, the training images are used to train the classifiers, and the testing images are classified and used to compute the different precision/recall values. The parameters are changed in PARAMS. For a given parameter, each possible parameter value you try (fo...
von diane
9. Jun 2010 18:12
Forum: Computer Vision
Thema: No functions provided in the new starting kit
Antworten: 1
Zugriffe: 412

Re: No functions provided in the new starting kit

1. is the correct answer, you don't need new glasses yet :) The 2 first parts were more guided, because we wanted everybody to reach the detection part. For the latter, we would like you to start from scratch (even if the structure of most functions is given in the task description so it is not real...
von diane
3. Jun 2010 13:14
Forum: Computer Vision
Thema: HOG cell visualization
Antworten: 2
Zugriffe: 542

Re: HOG cell visualization

This is a good remark. Contours and gradient vectors are suppose to be orthogonal. And that's the visualization we are using for the exercises. But in the case of slide 17, a different visualization scheme has been chosen: the gradients are presented rotated by 90°, so the contours appear in a more ...
von diane
3. Jun 2010 12:28
Forum: Computer Vision
Thema: Questions regarding the project
Antworten: 1
Zugriffe: 545

Re: Questions regarding the project

These are 2 very interesting questions which both needs quite lengthy explanations.
We prepared slides for tomorrow so we can answer in details.
von diane
26. Mai 2010 22:39
Forum: Computer Vision
Thema: Ex 3 Hessian threshold Q1 and Q4
Antworten: 1
Zugriffe: 529

Re: Ex 3 Hessian threshold Q1 and Q4

As explained in the answer for the topic "Ex3 1a)", there is different possible implementations of the second derivative, and the threshold needs to be adapted accordingly.
It does not mean that your solution is not correct, it means that you computed the second derivative in a different way.
von diane
26. Mai 2010 10:44
Forum: Computer Vision
Thema: EX3 3.) c.)
Antworten: 2
Zugriffe: 413

Re: EX3 3.) c.)

The co-occurrences of Mag values and Lap values (i.e. considering them as couples) are more informative than a simple concatenation of these values. I indeed expect that you use a 2D histogram (nb_bins X nb_bins) for computing the mag/lap histogram. As always, at the end of the function, the returne...
von diane
26. Mai 2010 10:32
Forum: Computer Vision
Thema: Ex3 1a)
Antworten: 3
Zugriffe: 405

Re: Ex3 1a)

My answer would be that in all cases you have an approximation of 1) a gaussian 2D smoothing, and 2) the computation of the second order derivative. Because everything is done in a discrete (and not continuous) space, with masks. The most common approximation, which also allows to precompute as much...
von diane
25. Mai 2010 10:18
Forum: Computer Vision
Thema: Ex3 1b
Antworten: 1
Zugriffe: 341

Re: Ex3 1b

The exercise sheet is indeed not so clear about this detail. For the Hessian function (hessian.m), the H variable shouldn't output the actual Hessian matrix, but the values of the Hessian determinant, which is the information necessary to extract the interest points, as it gives something similar to...
von diane
18. Mai 2010 19:53
Forum: Computer Vision
Thema: Allowed functions for exercise 2
Antworten: 1
Zugriffe: 458

Re: Allowed functions for exercise 2

As soon as you use them in an appropriate manner (not just applying histograms, but producing the exact required descriptor) I am fine with you using the built-in matlab functions.
von diane
17. Mai 2010 12:02
Forum: Computer Vision
Thema: EX2 1. b.)
Antworten: 1
Zugriffe: 487

Re: EX2 1. b.)

In exercise 2 - question 1, b), there are 3 different histograms to implement. I guess you are talking about the 2 different color histograms. I already mentioned (in other answers of this forum) that the output needs to be a normalized 1D histogram. The two different color histograms have indeed si...

Zur erweiterten Suche