AutoDDPM Implementation

Production-oriented implementation for single-subject autoDDPM inference, including:

• a managed CLI (./auto)
• clean inference entrypoint (inference_clean.py)
• model/runtime files needed for inference

What This Repo Focuses On

• Single-subject T1w anomaly detection using pretrained weights
• Slurm-friendly operations (start, stop, logs, checks)
• Minimal documentation and operational workflow

Quick Usage

cd /path/to/autoDDPM
./auto checks --input "/path/to/subject_T1w.nii.gz" --output "results_subject"
./auto start --input "/path/to/subject_T1w.nii.gz" --output "results_subject" --threshold -1
./auto logs --follow

See QUICK_START.md for full command details.

Threshold Guidance

• 0.13: paper-style fixed threshold
• -1: per-scan auto threshold (recommended when no healthy calibration set is available)

Reference

Method reference paper:

• https://arxiv.org/abs/2305.19643

Original upstream repository:

• https://github.com/ci-ber/autoDDPM