Lstm choice ril

[noahmetzger]

Overhaul of the lstm_choice_mode

• Made the lstm_choice_mode compatible with hocr_char_boxes.

• Improved the choice mode algorithm to increase the performance.

• Cut down the different modes to the two essentiell ones.

  • lstm_choice_mode=1 returns the timesteps segemented by character. This replaces mode 1 and 3 with no loss of information.
  • lstm_choice_mode=2 returns alternative choices per character based on the new ctc retrieval algorithm. This replaces mode 2 and 4. The former mode 2 is completely dropped as it worked only for languages with a smaller amount of characters than output channels of the LSTM.

• Renamed lstm_choice_amount to lstm_choice_iterations because of the misleading title.

[zdenop]

thanks

[woodjohndavid]

Unless I misunderstand what these options are supposed to do, it appears like there is a bug or oversight. Please refer to the user area thread:

https://groups.google.com/forum/#!topic/tesseract-ocr/5tC6appoUgE

There seems to be no way to prevent lstm from including duplicates in the generated text and/or HOCR output. The example in the thread above is a clear example of this.

Thanks.

[bertsky]

Unless I misunderstand what these options are supposed to do, it appears like there is a bug or oversight. Please refer to the user area thread:

https://groups.google.com/forum/#!topic/tesseract-ocr/5tC6appoUgE

@woodjohndavid the above change (and the preceding ones relating to this config variable) is not active unless you set lstm_choice_mode to 1 or 2 – the default is 0.

There seems to be no way to prevent lstm from including duplicates in the generated text and/or HOCR output. The example in the thread above is a clear example of this.

This is an independent problem, let's discuss it further in issue #2738 you created.