[GitHub] thomelane commented on a change in pull request #13647: Update lip reading example

[GitBox <gi...@apache.org>]

thomelane commented on a change in pull request #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#discussion_r246571676
 
 

 ##########
 File path: example/gluon/lipnet/BeamSearch.py
 ##########
 @@ -0,0 +1,168 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+"""
+Module : this module to decode using beam search
+https://github.com/ThomasDelteil/HandwrittenTextRecognition_MXNet/blob/master/utils/CTCDecoder/BeamSearch.py 
+"""
+
+from __future__ import division
+from __future__ import print_function
+import numpy as np
+
+class BeamEntry:
+    """
+    information about one single beam at specific time-step
+    """
+    def __init__(self):
+        self.prTotal = 0 # blank and non-blank
+        self.prNonBlank = 0 # non-blank
+        self.prBlank = 0 # blank
+        self.prText = 1 # LM score
+        self.lmApplied = False # flag if LM was already applied to this beam
+        self.labeling = () # beam-labeling
+
+class BeamState:
+    """
+    information about the beams at specific time-step
+    """
+    def __init__(self):
+        self.entries = {}
+        
+    def norm(self):
+        """
+        length-normalise LM score
+        """
+        for (k, _) in self.entries.items():
+            labelingLen = len(self.entries[k].labeling)
+            self.entries[k].prText = self.entries[k].prText ** (1.0 / (labelingLen if labelingLen else 1.0))
+
+    def sort(self):
+        """
+        return beam-labelings, sorted by probability
+        """
+        beams = [v for (_, v) in self.entries.items()]
+        sortedBeams = sorted(beams, reverse=True, key=lambda x: x.prTotal*x.prText)
+        return [x.labeling for x in sortedBeams]
+
+def applyLM(parentBeam, childBeam, classes, lm):
+    """
+    calculate LM score of child beam by taking score from parent beam and bigram probability of last two chars
+    """
+    if lm and not childBeam.lmApplied:
+        c1 = classes[parentBeam.labeling[-1] if parentBeam.labeling else classes.index(' ')] # first char
+        c2 = classes[childBeam.labeling[-1]] # second char
+        lmFactor = 0.01 # influence of language model
+        bigramProb = lm.getCharBigram(c1, c2) ** lmFactor # probability of seeing first and second char next to each other
+        childBeam.prText = parentBeam.prText * bigramProb # probability of char sequence
+        childBeam.lmApplied = True # only apply LM once per beam entry
+
+def addBeam(beamState, labeling):
+    """
+    add beam if it does not yet exist
+    """
+    if labeling not in beamState.entries:
+        beamState.entries[labeling] = BeamEntry()
+
+def ctcBeamSearch(mat, classes, lm, k, beamWidth):
 
 Review comment:
   Any chance you could add a quick unit test for this function? It looks complex, and could very easily contain a bug.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services