Skip to content

Advanced training

Warning

The recommended way to train models is through the napari training GUI. This section gives additional background information not needed for intended training procedure.

For training models, we rely on the pytorch-3dunet.
You can import models trained using this library into PlantSeg by choosing ADD CUSTOM MODEL in the model selection drop-down in the segmentation tab. Models are by default stored in ~/.plantseg_models/.

Each model needs the following files:

  • Configuration file used for training: config_train.yml
  • A snapshot of the best model across training: best_checkpoint.pytorch
  • A Snapshot of the last model saved during training: last_checkpoint.pytorch

Check the model directory to see examples of the config_train.yml