run_padis_training_smoke.py 🚧

Source: scripts/dev/run_padis_training_smoke.py

Warning

This file has not yet received a complete narrative and docstring audit. Its public source-level API is listed automatically below.

Smoke-test PaDIS training paths before launching expensive runs.

This script is intentionally small and destructive-free. It can load the real LIDC image-prior dataset, construct the real PaDIS NCSN++ model, and run a small number of optimizer steps with explicit patch sizes. Use CUDA only for tiny synthetic checks unless the target machine has enough VRAM for the full training batch.

LION_ROOT

run_smoke(args: argparse.Namespace) → dict[str, object]

Run a bounded PaDIS training/checkpoint smoke test.

build_parser() → argparse.ArgumentParser

Construct the training-smoke command-line parser.

main() → None

Execute the training smoke test and emit its JSON report.