examples

Virtual Stain Flow Examples

This directory contains example notebooks and scripts demonstrating how to use the virtual_stain_flow library for training image-to-image translation models.

Quick Start

1. Download Data

Download the JUMP pilot dataset from AWS S3 (public access, no credentials required):

python 0.download_data.py --outdir /YOUR/DATA/PATH/

This downloads the full 50GB of brightfield and fluorescence microscopy images for the default batch and plate hardcoded in the script.

2. Run Examples

Two example notebooks demonstrate core workflows:

• 1.modular_unet_example.ipynb - Building and configuring UNet models. Does not require dataset downloads.
• 2.training_with_logging_example.ipynb - Training with MLflow logging and callbacks. Requires dataset and setting up of a mlflow tracking server.

Requirements

See the project's pyproject.toml. Note that for data access, AWS cli is additionally required.

Data

Examples use the JUMP Pilot public dataset (CPJUMP1):

• Source: AWS S3 bucket (public access)
• Content: Multi-channel microscopy images (brightfield, Hoechst, GFP, etc.)
• Reference: JUMP Pilot Project