Restricted Boltzmann Machine - preprocessing data

Asked Mar 20 '15 at 08:55

Active Mar 20 '15 at 09:42

Viewed 427 times

I am programming on MATLAB and want to use RBMs with real-valued input, like greyscale images, so I tried to follow what Hinton said in this article.

The images have integer values in [0, 255] and are stored in a matrix D which is [numImages x numPixel]. So I started preprocessing the data:

scaled the entire dataset so that all the values are in [0, 1] with
```
D = D / 255;
```
brought every pixel to have zero mean across all images, so I subtracted from every column of the matrix its mean value with
```
imgMean = mean(D);    % row vector
D = D - repmat(imgMean, rows, 1);
```
divided the entire dataset by its standard deviation, so that every pixel has unit variance, with
```
D = D / std(D(:));
```

But when I try to plot the images, the result is clearly very dark, since many values become negative and are clipped to zero.

Is this ok or did I make any mistake with the preprocessing?

edited Mar 20 '15 at 09:42

asked Mar 20 '15 at 08:55

minomic

How are you plotting the images? – 3lectrologos Mar 20 '15 at 09:50
I am using the code provided by Hinton himself. Probably the interesting part about plotting is `imagesc(img2,[0 1]);` where img2 is what I want to plot, which is built as a "collage" of images from the test set and their reconstructions. – minomic Mar 20 '15 at 10:28
Does changing `imagesc` to `imshow` help? – 3lectrologos Mar 20 '15 at 14:18
Unfortunately no... The result is pretty much the same – minomic Mar 20 '15 at 14:32

0 Answers0