crnn
crnn copied to clipboard
How to support beam search decoding
I am following the pytorch implementation. I couldnt find any support for beam search decoding in either of the implementations. Does the decoding by default use beam search?
CRNN uses best path decoding: https://github.com/bgshih/crnn/blob/master/src/cpp/ctc.cpp#L152
is it better than a beam search with/without a language model?
best path decoding < beam search < beam search with (good) language model