As an example for available Identify Fibers (AI) options serves a sample of a scan of bulky nonwoven fabric.

To start, select Identify Fibers (AI) from the pull-down menu in the FiberFind section. Click the Options’ Edit... button.

The Fiber Identification dialog that opens contains four tabs: Fiber Identification, GPU Options, Output Options and Equations & References.

First, choose between two Identification Methods: Machine Learning for the AI approach, or Analytic to use the classical image processing approach. Note, that the Analytic method is deprecated and will be removed in future GeoDict versions.

For Machine Learning as the Identification Method, several options are available in the Fiber Identification dialog. However, for most of the application cases, the default values can be kept for most of the options.

Two choices are available as Identification Mode: Use Current Structure and Load Neural Network Output.

If you choose the default Use Current Structure, Identify Fibers (AI) will be applied to the structure that GeoDict currently has in memory.

Select the neural network listed under Neural Network (.gnn stands for GeoDict neural network). Besides the latest Neural Network provided with the software (LabelFibers.gnn), Neural Networks trained with GeoDict-AI can be used, too. For this, the corresponding .gnn-file has to be copied into the installation directory for Identify Fibers (AI) (e.g. C:\Program Files\Math2Market GmbH\GeoDict 2025\FiberFind\IdentifyFibers) in order to be selectable in the Neural Network field.

In the Description field, the current constraints for the application of the neural network are listed, such as the compatible fiber diameter range.

On occasion, you want to change the threshold without having to run the identification for each voxel with the neural network again, the option Load Neural Network Output may be very useful. Starting with a given Neural Network Output saves calculation time, as the Neural Network probability results can be used before new threshold-based calculations are made.

Browse to a GeoDict result folder of a previous run of Identify Fibers (AI) and select a nnOutput.npz file from that folder. The setting Material to Analyze must match the settings the Neural Network was trained with, and the loaded structure needs to be the same.

The Material to Analyze panel offers the choice of whether Identify Fibers (AI) should be applied to all solid and porous materials in the structure or only on a subset, defined either by choosing Material IDs or Materials.

The neural network provides probabilities about whether a solid voxel is a fiber centerline voxel or not. The Threshold parameter defines the limit for this labelling and can therefore be defined by the user between 0 and 1 (default 0.5). It is strongly recommended using a threshold higher than 0, since otherwise every voxel with solid material will be defined as the centerline of a fiber, and it becomes very hard to identify two fibers if they are connected.

Another aspect is that in the case of identified fibers are over-segmented, choosing a smaller threshold can lead to better results, see example in the Fiber identification visualization section.

Check Remove Fiber Fragments at Domain Boundary in the Domain-Boundary Options panel, to remove all fibers that touch the boundary of the domain after the fiber identification algorithm.

The underlying domain can be set as periodic for the fiber identification by checking one or several of the boxes in the Domain-Boundary Options panel. However, periodicity is unusual for most 3D scans and the default setting is rarely changed.

Finally, in the lower right of the Fiber Identification tab, you can select the type of fibers in the structure in the Fiber-Shape Analysis panel. Curved Circular Fiber or Short Circular Fiber are available as Fiber-Fit Shape.

Selecting Curved Circular Fiber, FiberFind performs the identification under the assumption that the structure might contain curved fibers.

If Short Circular Fiber is chosen, fibers are identified twice: once with the default option of curved fibers and once under the assumption of straight fibers only. For the computation of fiber length distribution and fiber orientation, always the result of the curved fiber identification is used.