1LSTMTRAINING(1)
2===============
3:doctype: manpage
4
5NAME
6----
7lstmtraining - Training program for LSTM-based networks.
8
9SYNOPSIS
10--------
11*lstmtraining*
12  --continue_from  'train_output_dir/continue_from_lang.lstm'
13  --old_traineddata 'bestdata_dir/continue_from_lang.traineddata'
14  --traineddata   'train_output_dir/lang/lang.traineddata'
15  --max_iterations 'NNN'
16  --debug_interval '0|-1'
17  --train_listfile 'train_output_dir/lang.training_files.txt'
18  --model_output  'train_output_dir/newlstmmodel'
19
20DESCRIPTION
21-----------
22lstmtraining(1)  trains LSTM-based networks using a list of lstmf files and starter traineddata file as the main input. Training from scratch is not recommended to be done by users. Finetuning (example command shown in synopsis above) or replacing a layer options can be used instead. Different options apply to different types of training.
23Read the [training documentation](https://tesseract-ocr.github.io/tessdoc/TrainingTesseract-4.00.html) for details.
24
25OPTIONS
26-------
27
28'--debug_interval  '::
29  How often to display the alignment.  (type:int default:0)
30
31'--net_mode  '::
32  Controls network behavior.  (type:int default:192)
33
34'--perfect_sample_delay  '::
35  How many imperfect samples between perfect ones.  (type:int default:0)
36
37'--max_image_MB  '::
38  Max memory to use for images.  (type:int default:6000)
39
40'--append_index  '::
41  Index in continue_from Network at which to attach the new network defined by net_spec  (type:int default:-1)
42
43'--max_iterations  '::
44  If set, exit after this many iterations. A negative value is interpreted as epochs, 0 means infinite iterations.  (type:int default:0)
45
46'--target_error_rate  '::
47  Final error rate in percent.  (type:double default:0.01)
48
49'--weight_range  '::
50  Range of initial random weights.  (type:double default:0.1)
51
52'--learning_rate  '::
53  Weight factor for new deltas.  (type:double default:0.001)
54
55'--momentum  '::
56  Decay factor for repeating deltas.  (type:double default:0.5)
57
58'--adam_beta  '::
59  Decay factor for repeating deltas.  (type:double default:0.999)
60
61'--stop_training  '::
62  Just convert the training model to a runtime model.  (type:bool default:false)
63
64'--convert_to_int  '::
65  Convert the recognition model to an integer model.  (type:bool default:false)
66
67'--sequential_training  '::
68  Use the training files sequentially instead of round-robin.  (type:bool default:false)
69
70'--debug_network  '::
71  Get info on distribution of weight values  (type:bool default:false)
72
73'--randomly_rotate  '::
74  Train OSD and randomly turn training samples upside-down  (type:bool default:false)
75
76'--net_spec  '::
77  Network specification  (type:string default:)
78
79'--continue_from  '::
80  Existing model to extend  (type:string default:)
81
82'--model_output  '::
83  Basename for output models  (type:string default:lstmtrain)
84
85'--train_listfile  '::
86  File listing training files in lstmf training format.  (type:string default:)
87
88'--eval_listfile  '::
89  File listing eval files in lstmf training format.  (type:string default:)
90
91'--traineddata  '::
92  Starter traineddata with combined Dawgs/Unicharset/Recoder for language model  (type:string default:)
93
94'--old_traineddata  '::
95  When changing the character set, this specifies the traineddata with the old character set that is to be replaced  (type:string default:)
96
97HISTORY
98-------
99lstmtraining(1) was first made available for tesseract4.00.00alpha.
100
101RESOURCES
102---------
103Main web site: <https://github.com/tesseract-ocr> +
104Information on training tesseract LSTM: <https://tesseract-ocr.github.io/tessdoc/TrainingTesseract-4.00.html>
105
106SEE ALSO
107--------
108tesseract(1)
109
110COPYING
111-------
112Copyright \(C) 2012 Google, Inc.
113Licensed under the Apache License, Version 2.0
114
115AUTHOR
116------
117The Tesseract OCR engine was written by Ray Smith and his research groups
118at Hewlett Packard (1985-1995) and Google (2006-present).
119