dwave-examples / image-segmentation Goto Github PK

The data type of the image data is uint8, which means that the difference calculations in the weight function will wrap around when negative values are encountered. This produces incorrect (and asymmetric) values for the differences between pixels.

For example, using the test_2_segments.png image data:

In [12]: img[40,40,:]
Out[12]: array([255, 255, 255], dtype=uint8)

In [13]: img[26,26,:]
Out[13]: array([0, 0, 0], dtype=uint8)

In [14]: img[40,40,:] - img[26,26,:]
Out[14]: array([255, 255, 255], dtype=uint8)

In [15]: img[26,26,:] - img[40,40,:]
Out[15]: array([1, 1, 1], dtype=uint8)

It would be helpful to set expectations on the runtimes of this example, those with preinstalled images are much longer than I would have guessed (compared to small random generated and our other DQM examples),
Also, running one preinstalled image, within recommended parameter limits, fails (I assume this was tested in the past so might need also an IDE issue).

Here's what I got on a few executions in Leap IDE:

Random generated:

• python image_segmentation.py for dimensions set to 2 and segments set to 2 ran quickly (under a minute)
• python image_segmentation.py for dimensions set to 3 and segments set to 2 ran quickly (under a minute)
• python image_segmentation.py for dimensions set to 6 and segments set to 2 and to 4 ran quickly (under a minute)

Preinstalled images:

• python image_segmentation.py test_2_segments.png for 2 segments took ~15 minutes
• python image_segmentation.py test_4_segments.png for 4 segments took ~15 minutes
• python image_segmentation.py test_image.jpeg for 5 segments aborted after about 20 minutes:

Leap IDE /workspace/image-segmentation $ python image_segmentation.py test_image.jpeg

Reading in your image...

        Enter number of segments > 5

Preparing DQM object...
Killed

• python image_segmentation.py test_image.jpeg for 2 segments aborted after about 20 minutes:

Leap IDE /workspace/image-segmentation $ python image_segmentation.py test_image.jpeg

Reading in your image...

        Enter number of segments > 2

Preparing DQM object...
Killed