image properties issue

The Tamura and Haralick properties are calculated in a few minutes and seconds respectively for about 10 images, but first order statistics runs forever.

Please check.

regards

rajeev

Hi Rajeev,

I found the problem. It's related to the weighted centroid feature. Can you switch this feature off? Then the calculation of the FirstOrderStatistics should be faster. The bug will be fixed in tomorrows nightly build and in the next 1.2.1 release of KNIME Image Processing.

Thank you for reporting this issue,

 

Christian

 

 

thanks Christian

Does it work without the weighted-centroid?
 

Weighted-centroid and MAD need to be left out. Then, it works!

mass displacement too is an issue!

Can you also check segment features and image segment features?

Hi,

a.) Can't reproduce the problems with MAD. Could you provide me an example image such that I can reproduce the problem?

b.) Mass Displacement is directly related to the problem with Weighted Centroid, so the problem should be fixed in the current nightly build.

c.) Segment Features / Image Segment Features: What should I check here? The above mentoined features share the code with Segment Features / Image Features. So if I fix the bug there, it should be fixed everywhere :-)

In general: we are currently implementing completly new nodes for the feature calculation, as the current code is very outdated and slow. In the next release, there will be way more feature available, which should be calculated pretty fast.

Thank you very much for your feedback!


Christian

 

If image segment features are used serially with each node calculating one set of features among Haralick, Tamura, segment geom and first order stats, previous columns are lost upon execution of subsequent node.

 

Thus only the last feature set is retained on using 4 image segment nodes serially! As a workaround 3 joins are required.

I don't know if I understand. (1) Couldn't you just calculate all features in one node and (2) I can't imagine a situation where you calculate features and then calculate features again based on the extracted bitmasks?

Get heap errors when I have 16 images that are segmented into more than 400 segments. So i used 4 nodes serially, each calculating a diff. feature set. One could also loop thru' all segments and use only 1 node.

See attached screenshot. Output from last segment properties node only contains it's results and not of onces before that. I have routinely split descriptor / feature calculations among several nodes in other domains and columns are generally appended! I dunno if anything's different here?

Can you increase your heap-space?  http://tech.knime.org/faq#q4_2.

Concerning your serial setting: I don't think this will work. Reason: If you calculate features from one image/labeling pair, the resulting table will consist of n entries, one row for each featurevector which was extracted from given a certain label in the labeling (one feature vector for each label). What would be the input of the next Image Segment Feature Node?! You see what I mean?

I think, if increasing your heap is no option or it's already very high (which would be interesting to know why... could you send me some example data + workflow?!), I would suggest that you use several image segment feature nodes, but all connected with the same input and afterwards you use some joiners.

Anyway, there will be a complete new implementation of the Feature/Descriptor Nodes in one or two months. The features will be implemented way more memory/cpu efficient. However, a serial setting won't be possible either ;-).