Parameters:
The number of input images is 30. As a pre-processing step, the input images are resized by a factor of 0.70.For unsupervised learning, we randomly start at 1 initializations. Then 10 iterations are carried out.

For the sketch model, 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 4 active basis templates. For local inhibition between templates, the minimum distance between two activated template is 0.40 times the size of template. Allowed template rotations: [-2, -1, 0, 1, 2]. Allowed image resolutions (relative): [0.80, 1.00, 1.20].

For the region model, the number of Gaussian component in the Gaussian mixture model is 10. Pixels within 10 pixels from the image boundary are deemed as background.

Learned templates.

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

Learned sketch templates for 4 entries of the codebook (after 10 iterations) (some entries may be empty):

Learned segmentation templates (shown in P instead of logP) for 4 entries of the codebook (after 10 iterations) (some entries may be empty):

Show sequences of iterations of templates:





numbers of elements: [40, 36, 38, 24].

Input images and their parsing results:

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):

Showing the segmentation: