Package index • geodl

Terrain Derivatives

Compute land surface parameters (LSPs) from a digital terrain model and build multi-band terrain visualizations, using either terra or torch.

makeSlope(): makeSlope
makeAspect(): makeAspect
makeCrv(): makeCrv
makeTPI(): makeTPI
makeTRI(): makeTRI
makeHillshade(): makeAspect
makeTerrainVisTerra(): makeTerrainVisTerra
makeTerrainVisTorch(): makeTerrainVisTerra

Chip Generation and Inspection

Generate masks and image/mask chips from larger rasters, list and describe chips, and visually check chips prior to training.

makeMasks(): makeMasks
makeChips(): makeChips
makeChipsMultiClass(): makeChipsMultiClass
makeChipsDF(): makeChipsDF
makeBoundaryDistance(): makeBoundaryDistance
describeChips(): describeChips
viewChips(): viewChips
addCoordNoise(): addCoordNoise

Dynamic Chip Generation

Generate chips on-the-fly from points and larger rasters during training instead of writing chips to disk beforehand.

makeDynamicChipsSF(): makeDynamicChipsSF
saveDynamicChips(): saveDynamicChips
checkDynamicChips(): checkDymamicChips

Datasets and DataLoaders

torch dataset subclasses for geospatial semantic segmentation and functions for inspecting a mini-batch.

defineSegDataSet(): defineSegDataSet
defineDynamicSegDataSet() defineDynmamicSegDataSet(): defineDynamicSegDataSet
describeBatch(): describeBatch
viewBatch(): viewBatch
viewBatchPreds(): viewBatchPreds
viewDist(): viewDist

Model Architectures

Configurable UNet-family semantic segmentation architectures as torch::nn_module() subclasses.

defineUNet(): defineUNet
defineUNet3p(): defineUNet3p
defineMobileUNet(): defineMobileUNet
defineResNetUNet(): defineResNetUNet
defineEfficientUNetB2(): defineEfficientUNetB2
defineDLab(): defineDLab
defineTerrainSeg(): defineTerrainSeg
countParams(): countParams

Loss Functions

Unified focal loss and regression loss implementations for luz.

defineUnifiedFocalLoss(): defineUnifiedFocalLoss
defineUnifiedFocalLossDS(): defineUnifiedFocalLossDS
defineRegressionLoss(): defineRegressionLoss
make_ds_weights(): make_ds_weights

Training Metrics

luz_metric() implementations for semantic segmentation.

luz_metric_overall_accuracy(): luz_metric_overall_accuracy
luz_metric_f1score(): luz_metric_f1score
luz_metric_recall(): luz_metric_recall
luz_metric_precision(): luz_metric_precision

Training Callbacks

luz_callback() implementations for staged training, EMA, and logging.

callback_save_model_state_dict(): callback_save_model_state_dict
callback_per_stage_optimizer(): callback_per_stage_optimizer
callback_freeze_stages(): callback_freeze_stages
callback_unfreeze_stages(): callback_unfreeze_stages
callback_ema(): callback_ema
callback_ds_weight_decay(): callback_ds_weight_decay
callback_reconfigure_loss(): callback_reconfigure_loss
callback_training_log(): callback_training_log

Reproducibility

Seed R and torch RNG streams for reproducible workflows.

setSeed(): setSeed

Prediction and Accuracy Assessment

Generate spatial predictions from a trained model and assess accuracy from rasters, dataloaders, or tabular reference/predicted values.

predictSpatial(): predictSpatial
assessDL(): assessDL
assessRaster(): assessRaster
assessPnts(): assessPnts

Training Diagnostics

plotTrainingSchedule(): plotTrainingSchedule