# Training artifacts — CueDetat pool/pocket detector

How `MASTER_POOL_MODEL.tflite` (the deployed detector) was produced.

- `cuedetat_pocket_detector_kaggle.py` / `.ipynb` — Kaggle re-train of the YOLOv8n
  pocket/pool detector. Inputs: Kaggle datasets `hereliesaz/cue-detat` and
  `diveshcrazy/pool-table-balls-classification`.
- `args.yaml` — Ultralytics training configuration.
- `best.pt` — trained PyTorch checkpoint (resumable / re-exportable).
- `best.onnx` — ONNX export.
- `best_float16.tflite` — FP16 TFLite export (the form shipped, concatenated into the master file).

License: AGPL-3.0 (Ultralytics-derived).