search

dgyoo / pa3

Recent image representation as PA3 of the computer vision class.
7 stars 0 forks source link

question about gaussian sampleRegnDesc #14

Open 20155004 opened 8 years ago

20155004 commented 8 years ago

% numGaussian:
% maxNumSrcIm:
% numSamplePerGaussian

plz give me some tips how to use them...

what's the meaning of below two terms? % numGaussian: Number of Gaussians
% numSamplePerGaussian: Rough number of region activations which belong to a single Gaussian.

Yoon-Jaeshine commented 8 years ago

20153395 Jaeshin Yoon.

You need to get activations for training GMM.

In this case, the number of activation you have to obtain is ('numGaussian*numSamplePerGaussian').

That is, the number of activation you have to obtain in each picture is same with numSamplePerGaussian. (Randomly).

  1. I mean, if you get activation from one picture in the multiscale way, you can get about 2000~3000 activations.
  2. And then, obtain 256(numSamplePerGaussian) activation randomly.
  3. Finally, get activations until the number of activation becomes ('numGaussian*numSamplePerGaussian')->256000activations. (it means that you should use 1000 pictures randomly).
20155004 commented 8 years ago

Thank you..