diff --git a/coco.py b/coco.py
index ff4b11a04bff5889290b8ee5af268055a6f90fcc..94b6cd18b068cb6233002709ad240cc93f4b84bc 100644
--- a/coco.py
+++ b/coco.py
@@ -51,8 +51,9 @@ ROOT_DIR = os.getcwd()
 # Path to trained weights file
 COCO_MODEL_PATH = os.path.join(ROOT_DIR, "mask_rcnn_coco.h5")
 
-# Directory to save logs and trained model
-MODEL_DIR = os.path.join(ROOT_DIR, "logs")
+# Directory to save logs and model checkpoints, if not provided
+# through the command line argument --logs
+DEFAULT_LOGS_DIR = os.path.join(ROOT_DIR, "logs")
 
 
 ############################################################
@@ -321,10 +322,15 @@ if __name__ == '__main__':
     parser.add_argument('--model', required=True,
                         metavar="/path/to/weights.h5",
                         help="Path to weights .h5 file or 'coco'")
+    parser.add_argument('--logs', required=False,
+                        default=DEFAULT_LOGS_DIR,
+                        metavar="/path/to/logs/",
+                        help='Directory to save logs and checkpoints. Defaults to logs/')
     args = parser.parse_args()
     print("Command: ", args.command)
     print("Model: ", args.model)
     print("Dataset: ", args.dataset)
+    print("Logs: ", args.logs)
 
     # Configurations
     if args.command == "train":
@@ -341,10 +347,10 @@ if __name__ == '__main__':
     # Create model
     if args.command == "train":
         model = modellib.MaskRCNN(mode="training", config=config,
-                                  model_dir=MODEL_DIR)
+                                  model_dir=args.logs)
     else:
         model = modellib.MaskRCNN(mode="inference", config=config,
-                                  model_dir=MODEL_DIR)
+                                  model_dir=args.logs)
 
     # Select weights file to load
     if args.model.lower() == "coco":