[Python] How does tf.nn.ctc_greedy_decoder generates output sequences in tensorflow?

Given the logits (output from the RNN/Lstm/Gru in time major format i.e. (maxTime, batchSize, numberofClasses)), how does ctc greedy decoder performs decoding to generate output sequence.

I found this "Performs greedy decoding on the logits given in input (best path)" on its webpage https://www.tensorflow.org/api_docs/python/tf/nn/ctc_greedy_decoder.

One possibility is to select output class with maximum value at each time step, collapse repetitions and generate corresponding output sequence. Is it, ctc greedy decoder doing here or something else? Explanation using an example will be very useful.

Continue reading...