Tesseract 4.0: VGSLSpecs

123 views
Skip to first unread message

Kay-Michael Würzner

unread,
Dec 16, 2016, 10:01:17 AM12/16/16
to tesseract-ocr
Dear @,

I played around with training the new LSTM mode. According to the documentation of the network specification (https://github.com/tesseract-ocr/tesseract/wiki/VGSLSpecs), the last number in the first tupel called 'depth' corresponds to the type of input (i.e. 1 ... grayscale, 3 ... color). However one o the given examples uses '48' in this position:

[1,1,0,48 Lbx256 O1c105]

Using an assumingly corrected specification

[1,48,0,1 Lbx256 O1c105]

causes serious runtime issues namely each iteration takes several minutes and huge amounts of memory are adressed. Any hints on what I am doing wrong here?

Many thanks in advance,
Kay

ShreeDevi Kumar

unread,
Dec 16, 2016, 11:36:02 AM12/16/16
to tesser...@googlegroups.com

Did you try out the commands as per the LSTM training tutorial?


--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-ocr+unsubscribe@googlegroups.com.
To post to this group, send email to tesser...@googlegroups.com.
Visit this group at https://groups.google.com/group/tesseract-ocr.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/167ff50a-2fbb-463c-b637-c051b2e9da82%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Kay-Michael Würzner

unread,
Dec 16, 2016, 12:28:20 PM12/16/16
to tesseract-ocr
Yes,  I did and in principle everything works like a charm which is great. What I want to accomplish now is some understanding: Why do I have to set a documented parameter in some undocumented way or to be more precise set this parameter to a value which conflicts with the documentation to make the whole process work?

Cheers,
Kay


On Friday, December 16, 2016 at 5:36:02 PM UTC+1, shree wrote:

Did you try out the commands as per the LSTM training tutorial?

On 16-Dec-2016 8:31 PM, "Kay-Michael Würzner" <wuer...@gmail.com> wrote:
Dear @,

I played around with training the new LSTM mode. According to the documentation of the network specification (https://github.com/tesseract-ocr/tesseract/wiki/VGSLSpecs), the last number in the first tupel called 'depth' corresponds to the type of input (i.e. 1 ... grayscale, 3 ... color). However one o the given examples uses '48' in this position:

[1,1,0,48 Lbx256 O1c105]

Using an assumingly corrected specification

[1,48,0,1 Lbx256 O1c105]

causes serious runtime issues namely each iteration takes several minutes and huge amounts of memory are adressed. Any hints on what I am doing wrong here?

Many thanks in advance,
Kay

--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.

ShreeDevi Kumar

unread,
Dec 16, 2016, 12:38:55 PM12/16/16
to tesser...@googlegroups.com, Ray Smith

+ Ray Smith


To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-ocr+unsubscribe@googlegroups.com.

To post to this group, send email to tesser...@googlegroups.com.
Visit this group at https://groups.google.com/group/tesseract-ocr.
Reply all
Reply to author
Forward
0 new messages