Tesseract 4: Shuffling training instances and unicharset compression at the same time?

[kolomiyets]

Hi,

I noticed that when training with unicharset compression (train_mode)training instances are used sequentially from one lstmf training file. This causes a local model convergence (for the current training font), whereas other fonts (training instances) were not used for training at all. Shuffling is possible with train_mode 16 but in my case I have to use unicharset compression which is  train_mode 64.


Is it possible to use unicharset compression and to shuffle training instances at the same time?


Many thanks in advance.

Cheers,
Alex



[ShreeDevi Kumar]

Please see https://github.com/tesseract-ocr/tesseract/wiki/TrainingTesseract-4.00

80 is the default. I think it means both 64 and 16 are applied.

train_mode

int

80

Flags from TrainingFlags in lstmrecognizer.h Possible values= 64 for Compress unicharset, 16 for round-robin training.

ShreeDevi
____________________________________________________________
भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com

--
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/f70b622a-2daa-43ab-ab08-c5f09d272030%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[kolomiyets]

Thanks!

On Friday, May 12, 2017 at 10:24:21 AM UTC+2, shree wrote:

Please see https://github.com/tesseract-ocr/tesseract/wiki/TrainingTesseract-4.00

80 is the default. I think it means both 64 and 16 are applied.

train_mode

int

80

Flags from TrainingFlags in lstmrecognizer.h Possible values= 64 for Compress unicharset, 16 for round-robin training.

ShreeDevi
____________________________________________________________
भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com

On Fri, May 12, 2017 at 1:46 PM, 'kolomiyets' via tesseract-ocr <tesser...@googlegroups.com> wrote:

Hi,

I noticed that when training with unicharset compression (train_mode)training instances are used sequentially from one lstmf training file. This causes a local model convergence (for the current training font), whereas other fonts (training instances) were not used for training at all. Shuffling is possible with train_mode 16 but in my case I have to use unicharset compression which is  train_mode 64.


Is it possible to use unicharset compression and to shuffle training instances at the same time?


Many thanks in advance.

Cheers,
Alex



--
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.