Expand YOLO training AOIs safely

2026-07-12 23:42:29 +02:00
parent 53cd38a5b2
commit 0f49c980ba
11 changed files with 310 additions and 7 deletions
@@ -116,6 +116,7 @@ def audit_feature_collection(payload: dict[str, Any], geod: Any, shape: Any) ->
        raise SystemExit("Evidence GeoJSON must be a FeatureCollection")
    features = payload.get("features") or []
    false_negative_ids: set[str] = set()
+    reference_ids: set[str] = set()
    false_negative_areas: list[float] = []
    matched_reference_areas: list[float] = []
    bucket_counts = {
@@ -141,11 +142,13 @@ def audit_feature_collection(payload: dict[str, Any], geod: Any, shape: Any) ->
            )
        area_m2 = abs(float(geod.geometry_area_perimeter(geometry)[0]))
        bucket = area_bucket(area_m2)
+        reference_id = stable_reference_id(feature, geometry)
+        reference_ids.add(reference_id)
        bucket_role = "false_negative" if role == "false_negative" else "matched_reference"
        bucket_counts[bucket][bucket_role] += 1
        bucket_counts[bucket]["total_reference"] += 1
        if role == "false_negative":
-            false_negative_ids.add(stable_reference_id(feature, geometry))
+            false_negative_ids.add(reference_id)
            false_negative_areas.append(area_m2)
        else:
            matched_reference_areas.append(area_m2)
@@ -157,6 +160,7 @@ def audit_feature_collection(payload: dict[str, Any], geod: Any, shape: Any) ->
    total_reference = len(false_negative_areas) + len(matched_reference_areas)
    return {
        "false_negative_ids": false_negative_ids,
+        "reference_ids": reference_ids,
        "false_negative_count": len(false_negative_areas),
        "matched_reference_count": len(matched_reference_areas),
        "total_reference_count": total_reference,
@@ -284,16 +288,31 @@ def run_audit(portfolio_args: list[str], output_dir: Path) -> tuple[Path, Path]:
    for sample_slug in sorted(expected_slugs):
        portfolio_rows = []
        false_negative_sets = []
+        reference_sets = []
        for label, _ in parsed_portfolios:
            raw = portfolio_samples[label][sample_slug]
            false_negative_sets.append(raw["false_negative_ids"])
+            reference_sets.append(raw["reference_ids"])
            portfolio_rows.append(
-                {key: value for key, value in raw.items() if key != "false_negative_ids"}
+                {
+                    key: value
+                    for key, value in raw.items()
+                    if key not in {"false_negative_ids", "reference_ids"}
+                }
+            )
+        if any(reference_ids != reference_sets[0] for reference_ids in reference_sets[1:]):
+            counts = ", ".join(
+                f"{label}={len(reference_ids)}"
+                for (label, _), reference_ids in zip(parsed_portfolios, reference_sets, strict=True)
+            )
+            raise SystemExit(
+                f"Sample {sample_slug} has different reference populations across portfolios ({counts})"
            )
        persistent_ids = sorted(set.intersection(*false_negative_sets))
        sample_reports.append(
            {
                "sample_slug": sample_slug,
+                "reference_population_count": len(reference_sets[0]),
                "persistent_false_negative_count": len(persistent_ids),
                "persistent_reference_ids": persistent_ids,
                "portfolios": portfolio_rows,