Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - hilden

Pages: [1]
1
CT, µCT and FIB-SEM / Re: Labeling for AI segmentation
« on: November 29, 2023, 03:29:05 PM »
Hi Erik,

  • It is possible by using a GeoPy script. Attached find a script that labels the given gray value ranges with the according IDs. To run the script, the gray value image should be loaded in the Image Processing dialog. From the main GeoDict window select Macro - Execute Macro/Script and browse for the script. Click on Parameters to assign material IDs to value ranges in a table. Afterwards click Run to generate an AI-label file (*.gld). Load this file in the Image Processing dialog by clicking Load Labels and browsing for the file. To learn more about automation by scripting in GeoDict, refer to the GeoPy scripting User Guide: https://www.geodict.com/fileadmin/UserGuide/GeoDict2023/Automation2023.pdf.
  • Good news for you: With the new GeoDict 2024 release coming soon, a new feature is available. The "Magic Brush" analyzes the scan on clusters recognizing edges. Then, labels can be created by simply selecting these clusters.


Best regards,
Janine

2
Hello Erik,

Yes, there is. In the 3D Image Processing dialog under the Image Processing tab unfold Image Alignment and then Rotation. Select Automatic Rotation and choose Plane from the Automation Mode pull-down menu. When clicking apply the image is automatically aligned with two major axes. If you want to see a preview of the automatic rotation first, you can just click Suggest - Plane and observe the rotation overlay in the 2D visualization of your scan on the right.

Best regards,
Janine

3
Analysis (-Find modules) / Re: Number of overlap points and angles
« on: August 31, 2023, 09:42:19 AM »
Hi Elise,

the number of overlap points you can obtain with MatDict -> GAD Objects Analysis -> Analyze Objects. The Results-Report tab in the Result Viewer shows the Number of Overlap Objects.

To analyze only one overlap object, you could for example right-click in the structure in an overlap region. The context menu offers the possibility to Select GAD objects. Use the first option to select all GAD-objects in this point. Then on the right open the GAD Object Selection tab from the side bar. Click Edit Objects and obtain the direction (orientation) from all of the fibers.

Finding all angles could be automated in a Python script. How exactly should the result look like? Something like a table with the overlap indices in one column and the angles for each of these overlap objects in the right column? If there are more than two fibers crossing, should each angle between every combination of the fibers be given? Or do you need only the orientations of all the fibers in the different points?

Best regards,
Janine

4
Analysis (-Find modules) / Re: GeoDict-AI
« on: January 23, 2023, 12:28:14 PM »
Dear Ziho,

Yes, this is possible. You only need to create the same folder structure as Create Training Data would generate.

  • Create an empty folder with a meaningful name.
  • Inside create the folder Train. If needed also a folder Test can be created for the Validate Performance feature.
  • Inside the Train folder create as many folders as you have training data, named as desired for example in counting numbers starting with 0 as is done by the Create Training Data feature.
  • Inside each of these subfolders place an input.gdt and an output.gdt file. The input.gdt must have only two IDs (ID00 Pore and ID01 Solid) and output.gdt three ( ID00 Pore, ID01 Grains and ID02 Binder)
I hope, my answer helped you.
Best regards, Janine

5
Dear Ziho,
  • The number of needed training structures depends on how difficult it is to identify the target material in the structures. For very simple cases or for testing purposes 10 training structures are good. For more difficult parameter combinations 100 or more training structures can be needed.
  • The parameter range depends on the structures that should be analyzed by the network later.

    I hope, my answer helped you.
Best regards, Janine

Pages: [1]