Code and data: (ZIP).
Run StartFromHere.m in Matlab. You can monitor intermediate results in the folder: output/.

Parameters:
The number of training images is 1. The size of each active basis template is 100 (width) by 100 (height) pixels. Maximum number of Gabor elements in each template is 40. For Gabor wavelets we use a scale of 0.70 and 16 quantized orientations within PI (in radian). Each Gabor element is allowed to move 3 pixels and rotate 1 orientation step(s) at most.

In total the codebook consists of 3 active basis templates. For unsupervised learning, we randomly start at 1 initializations. Then 10 iterations are carried out. In the encoding step, the activated templates need to have a SUM2 score of at least 40. For local inhibition between templates, the minimum distance between two activated template is 0.40 times the size of template. Allowed template rotations: [-16, -15, -14, -13, -12, -11, -10, -9, -8, -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]. Allowed image resolutions (relative): [0.80, 1.00, 1.20]. As a pre-processing step, the input images are resized by a factor of 0.50.

Learned templates.

Show sorted templates according to their log-lik (some entries may be empty):

Learned templates for 3 entries of the codebook (after 10 iterations) (some entries may be empty):

Show sequences of iterations of templates:




numbers of elements: [30, 27].

Traing images and their sketches:

Sketching the observed images by overlaying the activated templates on them:

Showing only the activated templates (with color):

Showing only the activated templates (with bounding boxes):

The adjusted BIC is 582.21.