tuandoan998 / handwritten-text-recognition Goto Github PK

How did you trained it , On CPU or GPU? And how long it take to train this model?

Could you give the details of the dataset used?
(Do you used entire IAM dataset??)

After training the model, when I am running prediction.py I am getting this bad marshal data issue. I am attaching the screenshot below.

Kindly help me resolve this issue.

Hi author, there is something wrong when I run Prediction.py or Evaluation.py. A few days ago, the code had been working well until this bug suddenly appeared from nowhere, I dind't change anything in the code. Hope you help to to define this problem.

Hi

I am trying to understand the annotations of IAM dataset in lines.txt file.
`
--- lines.txt ---------------------------------------------------------------

 iam database line information

 format: a01-000u-00 ok 154 19 408 746 1663 91 A|MOVE|to|stop|Mr.|Gaitskell|from

 a01-000u-00  -> line id for form a01-000u
 ok              -> result of word segmentation
                       ok: line is correctly segmented
                        err: segmentation of line has one or more errors

                    notice: if the line could not be properly segmented
                            the transcription and extraction of the whole
                            line should not be affected negatively

 154             -> graylevel to binarize line
 19              -> number of components for this line
 408 746 1663 91 -> bounding box around this line in x,y,w,h format

 A|MOVE|to|stop|Mr.|Gaitskell|from
                 -> transcription for this line. word tokens are separated
                    by the character |

  a01-000u-00 ok 154 19 408 746 1661 89 A|MOVE|to|stop|Mr.|Gaitskell|from
  a01-000u-01 ok 156 19 395 932 1850 105 nominating|any|more|Labour|life|Peers
  a01-000u-02 ok 157 16 408 1106 1986 105 is|to|be|made|at|a|meeting|of|Labour`

Can you please explain what is 154, 156 and how to get those values?

Can you also explain what is 19 and how to get those values?

@tuandoan998, I am trying to understand how you have pre-processed your images and labels.
What is
https://github.com/tuandoan998/HTR-for-IAM/blob/ffa2696a744e7c2256282a8eb7712290ad9f4f5e/ImageGenerator.py#L20
the purpose of i_len or input_length?

You have initialised it to 30 for word model.
https://github.com/tuandoan998/HTR-for-IAM/blob/ffa2696a744e7c2256282a8eb7712290ad9f4f5e/Parameter.py#L7

And this is the only place you have used it:
https://github.com/tuandoan998/HTR-for-IAM/blob/ffa2696a744e7c2256282a8eb7712290ad9f4f5e/ImageGenerator.py#L57

Why multiply with 30?