LSTMTRAINING(1)

📦
version 5.5.3 · 12 versions
, in opensuse-leap 16.0

💅 lstmtraining(1) 💅

okay bestie, welcome to lstmtraining(1) — literally the training program for LSTM-based networks, and yes she's iconic, yes she's a little extra, and yes we're gonna get through all of it together 🫶

✨ NAME ✨

lstmtraining — Training program for LSTM-based networks. that's it, that's the vibe 💖

🦋 SYNOPSIS 🦋

lstmtraining --continue_from  train_output_dir/continue_from_lang.lstm
--old_traineddata bestdata_dir/continue_from_lang.traineddata
--traineddata   train_output_dir/lang/lang.traineddata --max_iterations
NNN --debug_interval 0|-1 --train_listfile
train_output_dir/lang.training_files.txt --model_output
train_output_dir/newlstmmodel

🌸 DESCRIPTION 🌸

so lstmtraining(1) trains LSTM-based networks, and the main input situation is: a list of lstmf files 📁 plus a starter traineddata file. that's the whole main-character setup.

real talk though — training from scratch is not recommended to be done by users, no cap. like, don't. 🙈

what you can do instead is finetuning (the example command up in the synopsis is literally the finetuning example, iykyk) or replacing a layer via the options. and heads up: different options apply to different types of training, so not every flag is gonna slay in every scenario.

read the training documentation for the details, bestie: https://tesseract-ocr.github.io/tessdoc/TrainingTesseract-4.00.html

🎀 OPTIONS 🎀

  • --debug_interval — how often to display the alignment. (type: int default: 0)
  • --net_mode — controls network behavior. (type: int default: 192)
  • --perfect_sample_delay — how many imperfect samples between perfect ones. (type: int default: 0)
  • --max_image_MB — max memory to use for images. (type: int default: 6000)
  • --append_index — index in continue_from Network at which to attach the new network defined by net_spec. (type: int default: -1)
  • --max_iterations — if set, exit after this many iterations. a negative value is interpreted as epochs, 0 means infinite iterations. (type: int default: 0)
  • --target_error_rate — final error rate in percent. (type: double default: 0.01)
  • --weight_range — range of initial random weights. (type: double default: 0.1)
  • --learning_rate — weight factor for new deltas. (type: double default: 0.001)
  • --momentum — decay factor for repeating deltas. (type: double default: 0.5)
  • --adam_beta — decay factor for repeating deltas. (type: double default: 0.999)
  • --stop_training — just convert the training model to a runtime model. (type: bool default: false)
  • --convert_to_int — convert the recognition model to an integer model. (type: bool default: false)
  • --sequential_training — use the training files sequentially instead of round-robin. (type: bool default: false)
  • --debug_network — get info on distribution of weight values. (type: bool default: false)
  • --randomly_rotate — train OSD and randomly turn training samples upside-down. (type: bool default: false)
  • --net_spec — network specification. (type: string default: empty)
  • --continue_from — existing model to extend. (type: string default: empty)
  • --model_output — basename for output models. (type: string default: lstmtrain)
  • --train_listfile — file listing training files in lstmf training format. (type: string default: empty)
  • --eval_listfile — file listing eval files in lstmf training format. (type: string default: empty)
  • --traineddata — starter traineddata with combined Dawgs/Unicharset/Recoder for language model. (type: string default: empty)
  • --old_traineddata — when changing the character set, this specifies the traineddata with the old character set that is to be replaced. (type: string default: empty)

👑 HISTORY 👑

lstmtraining(1) was first made available for tesseract4.00.00alpha. she's been serving since day one, periodt 💅

🌈 RESOURCES 🌈

🔥 SEE ALSO 🔥

  • tesseract(1) — the whole reason we're here, go read her too 🫶

🍓 COPYING 🍓

Copyright (C) 2012 Google, Inc. Licensed under the Apache License, Version 2.0 — keep the license intact bestie, it's the law ✨

💜 AUTHOR 💜

The Tesseract OCR engine was written by Ray Smith and his research groups at Hewlett Packard (1985-1995) and Google (2006-2018). literal legends, no cap 👑

okay that's the whole page, you ate that 💅 this was the lstmtraining(1) page, dated 2026-09-28 — go train something iconic, mwah 🎀💖

📦 where this page comes from