Expand regional corpus and aerial finetuning controls

scripts/evaluate_belgium_building_candidate.py

@@ -52,6 +52,14 @@ def metrics(tp: int, fp: int, fn: int) -> dict[str, float | int]:
     }
 
 
+def scale_box(box: tuple[float, float, float, float], factor: float) -> tuple[float, float, float, float]:
+    """Scale a detector box around its center for calibration-only geometry correction."""
+    x1, y1, x2, y2 = box
+    cx, cy = (x1 + x2) / 2, (y1 + y2) / 2
+    half_width, half_height = (x2 - x1) * factor / 2, (y2 - y1) * factor / 2
+    return cx - half_width, cy - half_height, cx + half_width, cy + half_height
+
+
 def read_references(path: Path, width: int, height: int) -> list[tuple[float, float, float, float]]:
     boxes = []
     for line in path.read_text(encoding="utf-8").splitlines() if path.is_file() else []:
@@ -88,6 +96,7 @@ def main() -> int:
         default=1000,
         help="Maximum detections retained per tile; dense Belgian urban tiles exceed YOLO's default 300.",
     )
+    parser.add_argument("--box-scale", type=float, default=1.0)
     args = parser.parse_args()
 
     from ultralytics import YOLO
@@ -119,7 +128,7 @@ def main() -> int:
     for tile, result in zip(tiles, results, strict=True):
         height, width = result.orig_shape
         predictions = [
-            (tuple(map(float, box)), float(score))
+            (scale_box(tuple(map(float, box)), args.box_scale), float(score))
             for box, score in zip(result.boxes.xyxy.cpu().tolist(), result.boxes.conf.cpu().tolist(), strict=True)
         ]
         observations.append(
@@ -163,6 +172,7 @@ def main() -> int:
         "test_time_augmentation": args.augment,
         "inference_imgsz": args.imgsz,
         "max_detections_per_tile": args.max_det,
+        "box_scale": args.box_scale,
         "tile_count": len(tiles),
         "sweeps": sweeps,
     }