Bug fix image_size as dict

Model wasnt loading when using encoder.image_size as dict, the VisionEncoderDecoderModel class expects that is an array instead of a dictionary.

config.json

@@ -123,10 +123,7 @@
       "0": "LABEL_0",
       "1": "LABEL_1"
     },
-    "image_size": {
-      "height": 2560,
-      "width": 1920
-    },
+    "image_size": [2560, 1920],
     "initializer_range": 0.02,
     "is_decoder": false,
     "is_encoder_decoder": false,