Plan and Preprocess
March 24, 2026 ยท View on GitHub
This guide covers dataset fingerprint extraction, experiment planning, and preprocessing.
Recommended command
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:
- Extract a dataset fingerprint
- Create one or more nnU-Net configurations
- Preprocess the data for those configurations
The output is written into nnUNet_preprocessed/DatasetXXX_Name.
Useful options
- Use
--no_pbarin non-interactive environments. - Use
-d 1 2 3to process multiple datasets. - Use
-c 3d_fullresif you already know which configuration you want. - Use
-hto 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.jsonnnUNetPlans.json- preprocessed data folders for the created configurations
Next step
Continue to Train models.