Tutorials

The notebooks below are available in the notebooks/ directory of the GitHub repository. Run them locally to learn how to use the Datamint Python API across different scenarios.

Getting Started

• 01_upload_data.ipynb: Upload images, DICOMs, and other resources to a Datamint project.

• 02_explore_data.ipynb: Query and explore resources already in a project.

Annotations

• 01_upload_annotations.ipynb: Import and manage image-level and frame-level classification annotations.

• 02_geometry_annotations.ipynb: Upload bounding boxes, lines, and other geometry annotations.

Datasets

• 01_project_scoped_splits.ipynb: Assign project-scoped train/val/test splits, inspect split records, and replay historical snapshots.

• 02_patient_wise_splits.ipynb: Split datasets by patient to avoid data leakage between train and test sets.

• 03_build_dataset.ipynb: Build and configure a PyTorch dataset from a Datamint project.

• 04_volume_dataset.ipynb: Work with 3D volume datasets (NIfTI, DICOM series).

• 05_import_dataset.ipynb: Import an already-labeled dataset (COCO, Pascal VOC, or YOLO format) into a project in one call.

Experiment Tracking

• 01_mlflow_manual_logging.ipynb: Log models and experiments manually to MLflow via Datamint.

Deployment

• 01_deploy_registered_model.ipynb: Deploy a model registered in MLflow as a managed Datamint endpoint.

• 02_deploy_external_model.ipynb: Adapt and deploy an externally-trained model in Datamint.

• 03_validate_model.ipynb: Validate a model before promoting it to production.

• 04_predict_images_volumes_and_videos.ipynb: Run a 2D model on images, 3D volumes, and video frames using automatic prediction bridges.

• 05_local_docker_build.ipynb: Build and run a model image locally with podman as a pre-deployment sanity check, and predict against it.

End-to-End Examples

Complete workflows from data upload to deployment.

Each notebook below trains on a public dataset (BCCD, FracAtlas, BUSI, Synapse). The download-and-upload step for each one is a single call into datamint.examples, a helper module that downloads the raw dataset, uploads it to a Datamint project, and creates the matching annotations:

from datamint.examples import bccd_dataset

project = bccd_dataset.create(project_name="bccd_detection")

Other available modules: fracatlas_dataset, busi_dataset, synapse_dataset.

Slice-based (2D)

• 01_fracatlas_classification.ipynb: End-to-end classification pipeline on the FracAtlas dataset.

• 02_busi_segmentation.ipynb: Train a 2D segmentation model on the BUSI dataset with UNetPPTrainer, including custom model integration.

• 03_bccd_detection.ipynb: Object detection pipeline on the BCCD blood cell dataset.

Full 3D

• 01_synapse_unetrpp.ipynb: Volumetric segmentation on the Synapse dataset using UNETR++.

• 02_synapse_nnunet.ipynb: Volumetric segmentation on the Synapse dataset using nnUNet.

SAM (promptable segmentation)

• 01_deploy_sam_model.ipynb: Deploy SAM 3 and run promptable segmentation with text, point, and box prompts.