Deep Learning Segmenter parameters

Model source

Sets the type of model to be used. Each has a set of settings which need to be properly adjusted to get the best result.

–

ONNX or CZANN model

Option to use locally saved 2D ONNX-based Deep Learning models (semantic):

ONNX: any compatible ONNX model including ONNX zoo models which use ONNX opset version 12 and lower (opset 7).

Zeiss DL models: including CZMODEL and CZANN

–

arivis Cloud AI model store

Option to use a DL instance segmentation model from the list of downloaded models from arivis Cloud AI model store.

Model

Sets a DL model to be used. The operation will extract all necessary info from the loaded model and update the UI accordingly.

Note: To select an ONNX or CZANN model, click or paste the path in the text field

Note: To access the arivis Cloud AI model store, click . Download the models and click to refresh the list of models available.

Channels

Select the channels that will be used for analysis. Channels are available based on the selected model and its input shape

Outputs

Sets the output of the Operation.

Note: click on to select the desired output(s).

Normalization

Sets a normalization range to be applied on the data before inference.

–

Full range

Full range uses the intensity range of the input data pixel type, e.g. for 8bit it is [0,255].

–

Data range

Data range uses the minimum and maximum values of in-put data over all timepoints.

–

Data range per timepoint

Data range per timepoint, uses the minimum and maximum values of input data per timepoint.

–

Manual

Manual allows users to select any desired range using the picker to be used for all channels and timepoints.

Customize model settings

Select this checkbox to be able to modify specific model parameters. Otherwise, default values are used.

–

Min overlap

Sets a minimum voxel count for overlapping tiles used in the model. The default values are extracted from the loaded model and are zero if no value has been defined in the metadata.

–

Input range

Sets the intensity range. The default range is between 0 and 1. Change this range to match the trained model.

–

Prediction range

Sets the range to match the output of the trained model. The de-fault range is considered to be between 0 and 1.

Apply tile blending

Applies smoother transition between the results of two tiles and reduces the tiling artifacts.