(#3) Defense: Logging images generated by defense models

src/defense/models.py

@@ -101,3 +101,66 @@ def post_train(self):
 
     def custom_callbacks(self) -> List[keras.callbacks.Callback]:
         pass
+
+
+if __name__ == "__main__":
+    from utils.dataset import Mnist, Cifar10
+    import argparse
+
+    parser = argparse.ArgumentParser(
+        description="Logging images generated by defense models"
+    )
+
+    parser.add_argument(
+        "--dataset",
+        "-d",
+        metavar="DATASET",
+        type=str,
+        help="Dataset for testing",
+        required=True,
+        choices=["mnist", "cifar10"],
+    )
+
+    parser.add_argument(
+        "--defense",
+        "-f",
+        metavar="DEFENSE",
+        type=str,
+        help="Defense method",
+        required=True,
+        choices=["reformer", "denoiser", "motd"],
+    )
+
+    args = parser.parse_args()
+
+    if args.dataset == "mnist":
+        input_shape = (28, 28, 1)
+        _, test_set = Mnist().dataset()
+
+    else:
+        input_shape = (32, 32, 3)
+        _, test_set = Cifar10().dataset()
+
+    if args.defense == "reformer":
+        defense_model = Reformer(
+            f"defense_reformer_{args.dataset}", input_shape=input_shape
+        )
+    elif args.defense == "denoiser":
+        defense_model = Denoiser(
+            f"defense_denoiser_{args.dataset}", input_shape=input_shape
+        )
+    else:
+        defense_model = Motd(
+            f"defense_motd_{args.dataset}",
+            input_shape=input_shape,
+            dataset=args.dataset,
+        )
+
+    progress = keras.utils.Progbar(1)
+
+    with defense_model.tensorboard_file_writer().as_default():
+        for idx, (x, _) in enumerate(test_set.take(1)):
+            y = defense_model.predict(x)
+            tf.summary.image(f"{args.defense.upper()} result - {idx}", [x, y])
+
+            progress.add(1)