ctc_loss_calculator.cc Not a valid path

weinman / cnn_lstm_ctc_ocr

Tensorflow-based CNN+LSTM trained with CTC-loss for OCR

GNU General Public License v3.0

498 stars 170 forks source link

ctc_loss_calculator.cc Not a valid path #52

Closed Parshwa27 closed 3 years ago

Parshwa27 commented 5 years ago

While training OCR I got the following error a couple of times only: Error: W tensorflow/core/util/ctc/ctc_loss_calculator.cc:144] No valid path found I don't understand what it means. Can anyone help?

weinman commented 5 years ago

Either a learned probability for a ground truth label sequence has collapsed to zero or your ground truth label sequence is longer than the feature sequence. See https://stackoverflow.com/questions/45130184/ctc-loss-error-no-valid-path-found

Parshwa27 commented 5 years ago

Is this the reason that the model is omitting every double character? E.g. "ned" instead of "need".

weinman commented 4 years ago

@Parshwa27 Not likely. More likely is that the model has not yet learned to insert the CTC blank character needed to prevent the repeated characters from being merged during the beam search.