Add operator YOLO tile dataset exporter

scripts/README.md

@@ -302,6 +302,44 @@ the resulting `.pt` file like any other local model asset: verify preflight,
 run the real-data matrix and compare persisted QA/QC metrics before activating
 it as a useful default.
 
+When whole-image training does not improve QA/QC, export a tile-level dataset
+with overlapping raster windows:
+
+```bash
+docker exec -it geointel python3 /app/scripts/export_operator_yolo_tile_dataset.py \
+  --manifest-path /app/storage/operator-data/operator_samples_manifest.json \
+  --output-dir /app/storage/operator-data/yolo-building-tile-dataset \
+  --tile-size 192 \
+  --stride 96 \
+  --negative-keep-ratio 0.5 \
+  --val-samples turnhout \
+  --force
+```
+
+The tile exporter clips GRB building bounding boxes into each tile, writes
+YOLO labels beside each tile image, keeps a deterministic ratio of empty
+negative tiles, and records `yolo_tile_dataset_summary.json` with
+`positive_tile_count`, `negative_tile_count` and skipped negative tile counts.
+It remains operator tooling only: no provider fetch, no API mutation and no
+automatic model training.
+
+Train against the tile dataset by pointing the existing wrapper at the tile
+output directory:
+
+```bash
+docker exec \
+  -e OPERATOR_YOLO_DATASET_DIR=/app/storage/operator-data/yolo-building-tile-dataset \
+  -e YOLO_BASE_MODEL_PATH=/app/models/yolov8n.pt \
+  -e TRAIN_MODEL_OUTPUT_PATH=/app/models/geointel-building-tile-detector.pt \
+  -e TRAIN_EPOCHS=30 \
+  -e TRAIN_IMGSZ=256 \
+  -e TRAIN_BATCH=4 \
+  -e TRAIN_WORKERS=0 \
+  -e TRAIN_DEVICE=cpu \
+  -e PYTHON_BIN=python3 \
+  geointel bash /app/scripts/train_operator_yolo_detector.sh
+```
+
 Export calibration QA evidence for visual review:
 
 ```bash