diff --git a/README.ipynb b/README.ipynb
index 8af4116..322e181 100644
--- a/README.ipynb
+++ b/README.ipynb
@@ -17,7 +17,38 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "## Features"
+    "## Command Line Interface"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Batch Detection\n",
+    "\n",
+    "CamTrapML includes a Command Line Interface (CLI) for running batch detection. The interface is designed to follow that of the MegaDetector run_batch_detection.py script, currently with some caveats. Currently the CLI assumes reccursive directory search for images and does not support image_queue which means the MegaDetector run_batch_detection.py will likely process faster. \n",
+    "\n",
+    "    $ python -m camtrapml.scripts.batch_detection model_name input_dir output_json_path\n",
+    "\n",
+    "The options available for `model_name` are:\n",
+    "\n",
+    "* `md2`: MegaDetector v2\n",
+    "* `md3`: MegaDetector v3\n",
+    "* `md4`: MegaDetector v4.1\n",
+    "\n",
+    "Add `--output_relative_filenames` to output filenames relative to the `input_dir` in the output JSON.\n",
+    "\n",
+    "E.g.\n",
+    "\n",
+    "    $ python -m camtrapml.scripts.batch_detection --output_relative_filenames md4 ~/Survey2022 ~/Survey2022/md.4.1.0.json \n",
+    "\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Programatic Usage"
    ]
   },
   {
diff --git a/README.md b/README.md
index 64b9510..3b67365 100644
--- a/README.md
+++ b/README.md
@@ -6,7 +6,29 @@
 
     $ pip install camtrapml
 
-## Features
+## Command Line Interface
+
+### Batch Detection
+
+CamTrapML includes a Command Line Interface (CLI) for running batch detection. The interface is designed to follow that of the MegaDetector run_batch_detection.py script, currently with some caveats. Currently the CLI assumes reccursive directory search for images and does not support image_queue which means the MegaDetector run_batch_detection.py will likely process faster. 
+
+    $ python -m camtrapml.scripts.batch_detection model_name input_dir output_json_path
+
+The options available for `model_name` are:
+
+* `md2`: MegaDetector v2
+* `md3`: MegaDetector v3
+* `md4`: MegaDetector v4.1
+
+Add `--output_relative_filenames` to output filenames relative to the `input_dir` in the output JSON.
+
+E.g.
+
+    $ python -m camtrapml.scripts.batch_detection --output_relative_filenames md4 ~/Survey2022 ~/Survey2022/md.4.1.0.json 
+
+
+
+## Programatic Usage
 
 ### Loading Data
 
diff --git a/camtrapml/scripts/batch_detection.py b/camtrapml/scripts/batch_detection.py
new file mode 100644
index 0000000..e0d6015
--- /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 [md2, md3, md4]",
+    )
+    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", encoding="utf-8") as file_handle:
+        dump(results, file_handle, indent=2)
+
+
+if __name__ == "__main__":
+    batch_detection()