odet CLI

Install the package (uv pip install -e . or pip install oriented-det), then run subcommands via the odet entry point (see pyproject.toml).

Command	Role
train	Config-based training
train-multi-gpu	torchrun wrapper around oriented_det.cli.train
preds / metrics	Validation inference and offline metrics (tools/save_predictions.py)
lr-finder, stats, tile-dota, image-demo, viewer	Data and training utilities
playground-csv, playground-to-dota	Playground CSV / DOTA export
export-onnx	ONNX export (export.scripts.export_onnx)
export-tf	ONNX + Keras detect bundle (export.scripts.export_tf)
export-detect	Keras bundle from existing ONNX (export.scripts.build_faster_rcnn_savedmodel)
export-preds	Val inference via Keras bundle (export.scripts.save_predictions_tf)

TF export commands require pip install "oriented-det[export]" (or the export requirements). They call pure Python entrypoints — no Makefile. Default artifacts: ./odet_export/. See export/README.md.

Subcommands load implementations from tools/ (train, preds, tiling, …). Reusable inference, checkpoint, and collate helpers live in oriented_det/runtime/. See the main README.