diff --git a/camtrapml/scripts/batch_detection.py b/camtrapml/scripts/batch_detection.py
new file mode 100644
index 0000000..b164ad8
--- /dev/null
+++ b/camtrapml/scripts/batch_detection.py
@@ -0,0 +1,112 @@
+"""
+Command Line Utility for batch detection.
+"""
+
+from argparse import ArgumentParser
+from json import dump
+from datetime import datetime
+from tqdm import tqdm
+from camtrapml.dataset import ImageDataset
+from camtrapml.detection.models.megadetector import (
+    MegaDetectorV2,
+    MegaDetectorV3,
+    MegaDetectorV4_1,
+)
+
+
+def parse_args():
+    """
+    Parse command line arguments.
+    """
+
+    parser = ArgumentParser()
+    parser.add_argument(
+        "model",
+        type=str,
+        help="Detection model to utilise [md4, md5a, md5b]",
+    )
+    parser.add_argument(
+        "dataset_path",
+        type=str,
+        help="Path to directory containing the images",
+    )
+    parser.add_argument("output_path", type=str, help="Path to store the JSON output")
+
+    parser.add_argument("--output_relative_filenames", action="store_true")
+
+    return parser.parse_args()
+
+
+def get_model(model_name):
+    """
+    Load a detection model based on a short name.
+    """
+
+    if model_name == "md2":
+        return MegaDetectorV2()
+
+    if model_name == "md3":
+        return MegaDetectorV3()
+
+    if model_name == "md4":
+        return MegaDetectorV4_1()
+
+    raise ValueError(f"Unknown model {model_name}")
+
+
+def detection_to_json_types(detection):
+    """
+    Convert a detection to a JSON-compatible dictionary.
+    """
+    detection["conf"] = float(detection["conf"])
+    detection["bbox"] = [float(x) for x in detection["bbox"]]
+    detection["category"] = str(detection["category"])
+    return detection
+
+
+def batch_detection():
+    """
+    Run detection on a batch of images.
+    """
+
+    args = parse_args()
+
+    model = get_model(args.model)
+
+    print("Enumerating images...", end="")
+    dataset = ImageDataset(args.dataset_path)
+    image_paths = list(tqdm(dataset.enumerate_images()))
+    print(" Done")
+
+    results = {"images": []}
+
+    for image_path in tqdm(image_paths):
+        if args.output_relative_filenames:
+            output_image_path = str(image_path.relative_to(args.dataset_path))
+        else:
+            output_image_path = str(image_path)
+
+        prediction = model.detect(image_path)
+
+        results["images"].append(
+            {
+                "file": output_image_path,
+                "detections": [
+                    detection_to_json_types(detection) for detection in prediction
+                ],
+            }
+        )
+
+    results["detection_categories"] = {"1": "animal", "2": "person", "3": "vehicle"}
+
+    results["info"] = {
+        "detection_completion_time": str(datetime.now()),
+        "format_version": "1.0",
+    }
+
+    with open(args.output_path, "w") as file_handle:
+        dump(results, file_handle, indent=2)
+
+
+if __name__ == "__main__":
+    batch_detection()