KonfAI

KonfAI is a YAML-driven deep learning framework for medical imaging built on top of PyTorch.

At the low level, KonfAI is organized around three workflows:

  • TRAIN for fitting models from configuration files

  • PREDICTION for exporting model outputs to structured datasets

  • EVALUATION for computing metrics on saved predictions

On top of these workflows, KonfAI also provides KonfAI Apps: packaged workflows that can run locally or through a remote server.

The framework is especially useful when you want to:

  • keep experiments reproducible and easy to inspect

  • describe models, datasets, transforms, metrics, and schedulers in YAML

  • iterate on medical imaging workflows without rewriting orchestration code

  • package mature workflows into reusable apps

If you are new to the project, the fastest path is:

  1. read Quickstart

  2. start from one of the shipped examples

  3. come back to Core concepts when you want to adapt the YAML

If you want one concrete recommendation: start with examples/Segmentation and the konfai TRAIN command before looking at KonfAI Apps.

Choose your start

  • I want the fastest first successful run: Quickstart

  • I want to copy a working example before learning the internals: Examples

  • I want to understand how the YAML maps to Python objects: Core concepts

  • I want to package a stable workflow behind a cleaner interface: KonfAI Apps

For most new users, the best first stopping point is simple: get examples/Segmentation to run through konfai TRAIN and confirm that Checkpoints/ and Statistics/ are created.