Plan and Preprocess

March 24, 2026 ยท View on GitHub

This guide covers dataset fingerprint extraction, experiment planning, and preprocessing.

For a new dataset, use:

nnUNetv2_plan_and_preprocess -d DATASET_ID --verify_dataset_integrity

DATASET_ID is the numeric dataset identifier. --verify_dataset_integrity is recommended the first time you run the command.

What this does

The command performs three steps:

  1. Extract a dataset fingerprint
  2. Create one or more nnU-Net configurations
  3. Preprocess the data for those configurations

The output is written into nnUNet_preprocessed/DatasetXXX_Name.

Useful options

  • Use --no_pbar in non-interactive environments.
  • Use -d 1 2 3 to process multiple datasets.
  • Use -c 3d_fullres if you already know which configuration you want.
  • Use -h to inspect all options.

Split commands

If you need more control, you can run the steps individually:

nnUNetv2_extract_fingerprint -d DATASET_ID
nnUNetv2_plan_experiment -d DATASET_ID
nnUNetv2_preprocess -d DATASET_ID

What to inspect afterward

After preprocessing, the dataset folder in nnUNet_preprocessed contains:

  • dataset_fingerprint.json
  • nnUNetPlans.json
  • preprocessed data folders for the created configurations

Next step

Continue to Train models.