Interactive annotator allows only for a limited analysis of images

Dear All,

I am frequently using the interactive annotator for image analysis in order to manually select regions of interest. In some workflows however, I have difficulties to use the tool, since I often can only annotate the first images and than the annotator seems to hang. This means, I cannot annotate more images and it is impossible to go back to the previously annotated images. It is possible to switch between images with the table view, but the image is not updated in the large view window.

If I leave the annotator and execute it it will only generate the labels of the first annotated images and provide empty cells for the images that could not be annotated.

I thought this could be a memory issue but changing the memory policy has no impact.

Interestingly this does not happen for all workflows but seemed to be more prominent if larger labels are applied.

Has anyone observed this before and is there any solution?

Thanks in advance and kind regards

Stefan

Hi Stefan,

I'm sorry to hear that. We had some problems with the Interactive Annotator recently and we are looking into it as soon as possible. I hope we will have a solution in a few days.

I will keep you posted!

Christian

PS: By chance, could you provide us your knime.log?

Hi Stefan,

I fixed some problems regarding the Interactive Annotator. Can you double-check tomorrows nightly build and see if you still experience problems?
 

Again, sorry for the trouble!

 

Christian

Hi Christian,

I ccould not find any knime.log file. Do I have to activate recording of a log-file?

I have installed the nightly builds as a software repository already.  Will the interactive annotator be automatically update by the nightly build when I update the software?

Kind regards and very impressed again by the quick reply.

Stefan

Hi Stefan

If you update KNIME it will download the latest build from the update site, the node will be upgraded  when you restart KNIME after the update. To start the update click on "File -> Update KNIME".

You can find your log file by clicking on "View -> Open KNIME Log" or on your harddrive within the KNIME  workspace folder at the location: ".metadata/knime/knime.log".

Best,

Gabriel

 

Hi Gabriel,

can I update the specific node without updating from 3.1 to 3.2. The update is for a version installed on our server. I am concerned that I may have to fix or deal with other extension after an update from 3.1 to 3.2 (as it happended for my desktop version).

Stefan

Hi Stefan,

Nodes can't be updated individually unfortunately, but you can update on plugin-basis (i.e. just KNIME Image Processing).

Cheers,

Christian

Hi Christian,

I was unable to test the nightly build. After updating to the nightly build the image reader was no longer working. Hence, I had to reinstall the image analysis nodes after removing the nightly build from the software repository. One temporary workaround we are using at present is when the annotator starts to hang, we make a copy, connect it to the previous node. Then we are able to continue the annotation.

Kind regards

Stefan

Hi Stefan,

we just published a new release of KNIP (1.5.2) which is available on the trusted community contributions update-site. You can update KNIME as described by Gabriel.

Does this work for your?

Best,

Christian

Hi All,

Interactive annotator is an interesting beast. I like its integration with KNIME but I don't understand why its so slow. I processed 100 large images by creating one annotation/image. It looks like annotator is copying the entire image set when I hit execute. I tried append vs new table and that makes no difference. I'd like to increase then number of images to process but that fails. 

Are there ways to annotate that are less computational intense or do I just have the settings wrong?

Hi Azijlstra,

I am happy you like the Interactive Annotator, is unfortunate that are experiencing this problem. Can you please give me more details on your images so that we can reproduce and investigate this issue?

The information that would be helpful can be found in the image properties section of the Image Viewer view: Type, Image type, Size X, Size Y, ...

The operation performed by the annotator should not be very costly, it looks like you might have disovered a bug.

best,

Gabriel

PS:

You can experiment with the options in the  Label Settings tab. I was able to speed up the processing by setting the Storage Img Type to BYTETYPE. Depending on your images you can also experiment with the Factory Type.

Hi Gabriel,

I'm using:

3000X3000 pixel JPG2000 RGB image with 3 channels approximately 5mb in size

In the label settings I've selected ntree_img_factory and the unsignedbytetype. "unique segment ID as label" is selected.

The original file of 5mb is stored (in filestore of the annotator) at 50mb. 

Thanks

Andries

HI Andries,

If the number of labeled points per image + the number of different labels is  than 256 you can try using the the setting BYTETYPE. In any case you should be safe to switch to the FactoryType ARRAY_IMG_FACTORY. Especially the first option should drastically improve the performance.

The original file of 5mb is stored (in filestore of the annotator) at 50mb.

This is because of the Factory Type (ShortType Images are much larger), as well as the fact that internal image files are saved uncompressed.