Bound evaluator source batches

scripts/evaluate_belgium_building_candidate.py

@@ -180,6 +180,8 @@ def main() -> int:
     parser.add_argument("--proposal-crop-scale", type=float, default=1.4)
     parser.add_argument("--proposal-classifier-batch", type=int, default=64)
     args = parser.parse_args()
+    if args.batch < 1:
+        parser.error("--batch must be positive")
     if args.proposal_classifier_batch < 1:
         parser.error("--proposal-classifier-batch must be positive")
     if not 0.0 < args.nms_iou < 1.0:
@@ -211,23 +213,28 @@ def main() -> int:
     }
     tiles = [item for item in summary["tiles"] if item.get("kept", True) and item["split"] == args.split]
     image_paths = [item["image_path"] for item in tiles]
-    results = YOLO(str(args.model)).predict(
-        image_paths,
-        conf=min(args.thresholds),
-        device=args.device,
-        augment=args.augment,
-        imgsz=args.imgsz,
-        batch=args.batch,
-        max_det=args.max_det,
-        iou=args.nms_iou,
-        verbose=False,
-    )
+    def predict_bounded(model: YOLO) -> list[Any]:
+        bounded_results: list[Any] = []
+        for start in range(0, len(image_paths), args.batch):
+            bounded_results.extend(
+                model.predict(
+                    image_paths[start : start + args.batch],
+                    conf=min(args.thresholds),
+                    device=args.device,
+                    augment=args.augment,
+                    imgsz=args.imgsz,
+                    batch=args.batch,
+                    max_det=args.max_det,
+                    iou=args.nms_iou,
+                    verbose=False,
+                )
+            )
+        return bounded_results
+
+    results = predict_bounded(YOLO(str(args.model)))
     additional_results = None
     if args.additional_model:
-        additional_results = YOLO(str(args.additional_model)).predict(
-            image_paths, conf=min(args.thresholds), device=args.device, augment=args.augment,
-            imgsz=args.imgsz, batch=args.batch, max_det=args.max_det, iou=args.nms_iou, verbose=False,
-        )
+        additional_results = predict_bounded(YOLO(str(args.additional_model)))
     observations: list[dict[str, Any]] = []
     for result_index, (tile, result) in enumerate(zip(tiles, results, strict=True)):
         height, width = result.orig_shape