LSTM: Training - Error in boxCreate: x < 0 and box off +quad

[Shreeshrii]

Extracting font properties only
Rendered page 242 to file /tmp/tmp.lswGVDuVeX/bih/bih.Mangal.exp0.tif
Rendered page 244 to file /tmp/tmp.lswGVDuVeX/bih/bih.Sanskrit_2003.exp0.tif
Rendered page 239 to file /tmp/tmp.lswGVDuVeX/bih/bih.Siddhanta.exp0.tif
Error in boxCreate: x < 0 and box off +quad
Rendered page 243 to file /tmp/tmp.lswGVDuVeX/bih/bih.Mangal.exp0.tif
Stripped 1 unrenderable words
Error in boxCreate: x < 0 and box off +quad
Rendered page 245 to file /tmp/tmp.lswGVDuVeX/bih/bih.Sanskrit_2003.exp0.tif
Error in boxCreate: x < 0 and box off +quad
Rendered page 240 to file /tmp/tmp.lswGVDuVeX/bih/bih.Siddhanta.exp0.tif
Rendered page 244 to file /tmp/tmp.lswGVDuVeX/bih/bih.Mangal.exp0.tif
Rendered page 246 to file /tmp/tmp.lswGVDuVeX/bih/bih.Sanskrit_2003.exp0.tif
Error in boxCreate: x < 0 and box off +quad
Error in boxCreate: x < 0 and box off +quad
Error in boxCreate: x < 0 and box off +quad
Error in boxCreate: x < 0 and box off +quad
Error in boxCreate: x < 0 and box off +quad
Rendered page 241 to file /tmp/tmp.lswGVDuVeX/bih/bih.Siddhanta.exp0.tif
Error in boxCreate: x < 0 and box off +quad
Error in boxCreate: x < 0 and box off +quad
Error in boxCreate: x < 0 and box off +quad
Rendered page 245 to file /tmp/tmp.lswGVDuVeX/bih/bih.Mangal.exp0.tif
Error in boxCreate: x < 0 and box off +quad
Rendered page 247 to file /tmp/tmp.lswGVDuVeX/bih/bih.Sanskrit_2003.exp0.tif
Rendered page 242 to file /tmp/tmp.lswGVDuVeX/bih/bih.Siddhanta.exp0.tif
Rendered page 246 to file /tmp/tmp.lswGVDuVeX/bih/bih.Mangal.exp0.tif
Rendered page 248 to file /tmp/tmp.lswGVDuVeX/bih/bih.Sanskrit_2003.exp0.tif
Error in boxCreate: x < 0 and box off +quad
Rendered page 243 to file /tmp/tmp.lswGVDuVeX/bih/bih.Siddhanta.exp0.tif
Rendered page 247 to file /tmp/tmp.lswGVDuVeX/bih/bih.Mangal.exp0.tif
Rendered page 249 to file /tmp/tmp.lswGVDuVeX/bih/bih.Sanskrit_2003.exp0.tif

[Shreeshrii]

Line 179
https://github.com/DanBloomberg/leptonica/blob/8a839a254e8d2ed3965ab0ecce7120b05f291993/src/boxbasic.c

[amitdo]

Extracting font properties only

I'm not sure if the LSTM engine needs this step.

[Shreeshrii]

I am using tesstrain.sh . The ngram file creation, which is used for font properties is done even before creation of images.

--linedata_only is the option used for creating the lstmf files.

training/tesstrain.sh --fonts_dir /home/shree/.fonts --lang bih  --linedata_only \
  --langdata_dir ../langdata --tessdata_dir ./tessdata \
  --output_dir ~/tesstutorial/bihnew

[amitdo]

I am using tesstrain.sh . The ngram file creation, which is used for font properties is done even before creation of images.

Yes, I know...

I think the LSTM engine does not need this info. If that's true, it should not activated in tesstrain.sh when training witn the --linedata_only option.

[Shreeshrii]

           --noextract_font_properties)
                EXTRACT_FONT_PROPERTIES=0 ;;

So, for LSTM training we should use both of these -

--linedata_only --noextract_font_properties

[amitdo]

Yes, probably... :-)

[theraysmith]

Correct. Changing instructions now...