Update per review

Change default value for output_blob to None
Move preprocess script and update link in README

Signed-off-by: Junze Wu <[email protected]>

Author: jnzw

demos/handwritten_text_recognition_demo/python/handwritten_text_recognition_demo.py

@@ -48,7 +48,7 @@ def build_argparser():
                       help="Path to the decoding char list file. Default is for Japanese")
     args.add_argument("-dc", "--designated_characters", type=str, default=None, help="Optional. Path to the designated character file")
     args.add_argument("-tk", "--top_k", type=int, default=20, help="Optional. Top k steps in looking up the decoded character, until a designated one is found")
-    args.add_argument("-ob", "--output_blob", type=str, default="output", help="Optional. Name of the output layer of the model. Default is \'output\'")
+    args.add_argument("-ob", "--output_blob", type=str, default=None, help="Optional. Name of the output layer of the model. Default is None, in which case the demo will read the output name from the model, assuming there is only 1 output layer")
     return parser
 
 
@@ -85,9 +85,13 @@ def main():
     net = ie.read_network(args.model, os.path.splitext(args.model)[0] + ".bin")
 
     assert len(net.input_info) == 1, "Demo supports only single input topologies"
-
     input_blob = next(iter(net.input_info))
-    out_blob = args.output_blob
+
+    if args.output_blob is not None:
+        out_blob = args.output_blob
+    else:
+        assert len(net.outputs) == 1, "Demo supports only single output topologies"
+        out_blob = next(iter(net.output_info))
 
     characters = get_characters(args)
     codec = CTCCodec(characters, args.designated_characters, args.top_k)

models/intel/handwritten-english-recognition-0001/README.md

@@ -18,7 +18,7 @@ The network is able to recognize English text consisting of characters in the [G
 | Accuracy on GNHK test subset (excluding images wider than 2000px after resized to height 96px with aspect ratio) | 81.5%     |
 | Source framework          | PyTorch\* |
 
-> **Note:** to achieve the accuracy, images from the GNHK test set should be binarized using adaptive thresholding, and preprocessed into single-line text images, using the coordinates from the accompanying JSON annotation files in the GNHK dataset. See [preprocess_gnhk.py](../../../demos/handwritten_text_recognition_demo/python/utils/preprocess_gnhk.py).
+> **Note:** to achieve the accuracy, images from the GNHK test set should be binarized using adaptive thresholding, and preprocessed into single-line text images, using the coordinates from the accompanying JSON annotation files in the GNHK dataset. See [preprocess_gnhk.py](./preprocess_gnhk.py).
 
 This model adopts [label error rate](https://dl.acm.org/doi/abs/10.1145/1143844.1143891) as the metric for accuracy.