Add leak-free regional YOLO expert datasets

scripts/build_regional_yolo_expert_dataset.py

@@ -0,0 +1,87 @@
+#!/usr/bin/env python3
+"""Create a leak-free train list for a regional YOLO expert."""
+
+from __future__ import annotations
+
+import argparse
+import hashlib
+import json
+from pathlib import Path
+
+
+def sha256(path: Path) -> str:
+    digest = hashlib.sha256()
+    with path.open("rb") as stream:
+        for chunk in iter(lambda: stream.read(1024 * 1024), b""):
+            digest.update(chunk)
+    return digest.hexdigest()
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--summary", type=Path, required=True)
+    parser.add_argument("--corpus-manifest", type=Path, required=True)
+    parser.add_argument("--region", required=True)
+    parser.add_argument("--output-dir", type=Path, required=True)
+    parser.add_argument("--positive-repeat", type=int, default=2)
+    parser.add_argument("--negative-repeat", type=int, default=1)
+    args = parser.parse_args()
+    if args.positive_repeat < 1 or args.negative_repeat < 1:
+        raise SystemExit("repeat factors must be positive")
+
+    summary = json.loads(args.summary.read_text(encoding="utf-8"))
+    manifest = json.loads(args.corpus_manifest.read_text(encoding="utf-8"))
+    samples = {item["sample_slug"]: item for item in manifest["samples"]}
+    paths: list[str] = []
+    selected_samples: set[str] = set()
+    positive_tiles = negative_tiles = 0
+    for tile in summary["tiles"]:
+        sample = samples[tile["sample_slug"]]
+        if sample["split"] != "train" or tile["split"] != "train":
+            continue
+        if sample["region"] != args.region:
+            continue
+        positive = int(tile.get("label_count") or 0) > 0
+        repeat = args.positive_repeat if positive else args.negative_repeat
+        paths.extend([str(Path(tile["image_path"]).resolve())] * repeat)
+        selected_samples.add(tile["sample_slug"])
+        positive_tiles += int(positive)
+        negative_tiles += int(not positive)
+    if not paths or not positive_tiles:
+        raise SystemExit(f"no positive train tiles found for region {args.region!r}")
+
+    args.output_dir.mkdir(parents=True, exist_ok=True)
+    train_list = args.output_dir / "train.txt"
+    train_list.write_text("\n".join(paths) + "\n", encoding="utf-8")
+    dataset_yaml = args.output_dir / "dataset.yaml"
+    dataset_yaml.write_text(
+        f"path: {args.output_dir}\ntrain: {train_list}\n"
+        f"val: {args.summary.parent / 'images' / 'val'}\nnames:\n  0: building\n",
+        encoding="utf-8",
+    )
+    evidence = {
+        "schema_version": 1,
+        "status": "ok",
+        "region": args.region,
+        "summary": str(args.summary),
+        "summary_sha256": sha256(args.summary),
+        "corpus_manifest": str(args.corpus_manifest),
+        "corpus_manifest_sha256": sha256(args.corpus_manifest),
+        "positive_repeat": args.positive_repeat,
+        "negative_repeat": args.negative_repeat,
+        "source_positive_tile_count": positive_tiles,
+        "source_negative_tile_count": negative_tiles,
+        "sampled_train_entry_count": len(paths),
+        "selected_train_samples": sorted(selected_samples),
+        "protected_samples_in_training": [],
+        "train_list": str(train_list),
+        "dataset_yaml": str(dataset_yaml),
+    }
+    evidence_path = args.output_dir / "regional-expert-dataset.json"
+    evidence_path.write_text(json.dumps(evidence, indent=2), encoding="utf-8")
+    print(json.dumps(evidence, indent=2))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())