Settings & Features

Basic Training Parameters

Parameter	Description	Default	Type
Raster Pairs	Input/target raster pairs for training. Each pair consists of an input raster and expected output raster. Use the Eval Only flag for a pair of rasters to exclude the rasters from training and only use them to evaluate the effectivness of the model on unseen data after training ends.	N/A	List of (input, target, evalOnly)
Number of Epochs	Number of complete passes through the training dataset.	10	Integer
Learning Rate	Controls how much to adjust model weights during optimization.	0.001	Float
Training Type	Type of task: classification (categorical output) or regression (continuous values).	Classification	Enum
Input Model	Existing model to continue training (optional).	None	File
Output Model	Location to save the trained model.	N/A	File

Parameter	Description	Default	Command Line Flag
Depth	Depth of the UNet model (number of down/up sampling operations). Higher depth helps with learning more complex relationships but increases training time	4	–depth, -d
Channels	Number of base feature channels in the first UNet layer. Higher values can help with learning more complex relationships but increases training time	64	–channels, -c

Parameter	Description	Default	Command Line Flag
NODATA Value	Value to treat as no data in input rasters. This is in addition to the Raster’s own NODATA value which is converted to this value before training.	-100	N/A
Normalize Input	Whether to normalize input values to [0,1] range.	True	N/A
Normalize Targets	Whether to normalize target values (regression only). Note: values will be denormalized during prediction if this is ‘True’	True	–normalize_targets, -n
Chunk Size	Size of image chunks used for training (e.g., 256×256 pixels).	256	–chunk_size, -ch
Rescale Factor	Downscale images by this factor (e.g., 0.5 = half size). Downscaling can increase training speed by reducing the number of chunks.	1.0	–rescale, -r

Parameter	Description	Default	Command Line Flag
Batch Size	Number of samples processed before model weights are updated.	16	–batch_size, -b
Validation Split	Fraction of data used for validation during training.	0.2	–validation_split, -v
Class Weights	Weighting for different classes (classification only). Useful for ignoring multiple classes, or increasing training sensitivity to classes with small sample sizes.	None	–weights, -w
Early Stopping Patience	Number of epochs with no improvement before training stops.	5	–end_patience, -ep
Save Mode	Whether to save best model (0) or last model (1).	0	–save_mode, -sm

Parameter	Description	Default	Command Line Flag
Profiling	Enable performance profiling during training. Results will be saved to plugins/QLearn/profile_results	False	–profile, -p

When using a trained model for prediction, the following parameters are available:

Parameter	Description	Default
Input Raster	Raster to perform prediction on.	N/A
Model	Trained model file (.pth) to use for prediction.	N/A
Output Raster	Location to save the prediction results.	N/A
Confidence Level	The minimum confidence level at which to output a prediction (classification only). Predictions below this confidence will be overwritten with the NODATA value supplied during training	0.0