Skip to content

Latest commit

 

History

History
214 lines (170 loc) · 7.19 KB

CHANGELOG.md

File metadata and controls

214 lines (170 loc) · 7.19 KB
Raw

Change Log

2.6.3

  • bugfix T2I: also ignore small regions in T2I, which are also ignored in HTR-jobs

2.6.2

  • make parent 1.8

2.6.1

  • bugfix T2I: use all transcripts of pageXML (instead of only special T2I-regions) when no transcripts are given

2.6.0

  • feature LA: if regions overlap, baselines are will not be separated or deleted.
  • feature HTR(x): If the process gets as property "use_lr" = "true", a language resource in the HTR folder is assumed. If "use_lr" = , this language resource is taken.

2.5.1

  • better error handling/notification/logging for OutOfMemoryException

2.5.0

  • feature T2I: if no file to a text file is given, method utilizes the special reference text lines in XML-file
  • feature T2I: new property: T2I_IGNORE_LB = "ig_lb": if "true", line breaks are substituted by spaces.
  • feature LA: leaves special reference text lines in XML-file.
  • feature LA: if property LA_DELETESCHEME = "la_deletescheme" is set to DEL_ALL = "all", also the special reference text lines are deleted

2.4.3

  • bugfix for HTR+ ignore very small line regions

2.4.2

  • bugfix for HTR+ delete old *.pb in export folder

2.4.1

  • bugfix for HTR+ base model training
  • needs tf_htsr version 3.0.5
  • training snippets are greyscale to reduce disc storage

2.4.0

  • make planet and private dependencies 'provided'
  • make dependent on new tokenizer
  • bugfix in Text2Image is used

2.3.2

  • bugfix format characters in transcripts: ignore lines and characters, when they are from an unknown category and report to observer.
  • bugfix T2I: do not delete folder, if input and ouput is the same

2.3.1

  • bugfix B2P: ignores text lines while B2P, if they have invalid baselines.

2.3.0

  • bugfix training: charmap cannot be in htrOut
  • bugfix language model: exception for empty transcript caught and reported via ErrorNotification
  • feature: slightly better preproc

2.2.2

  • bugfix training: shuffle training list so that order is "randomly"
  • bugfix surrogates: if a line contains surrogates, it is ignored in training and validation
  • bugfix empty validation: If a validation set is given, but no lines are in, throw exception
  • feature notification: errors while creating training/validation data are reported via notification

2.2.1

  • bugfix LA Advanced: when region is given and rotated text lines are found, lines will now not rotated out of image

2.2.0

  • update to newer versions for bugfix in planets sources

2.1.1

  • bugfix training of HTR+ with base model

2.1.0

  • faster TextAlignment and base technology moved to CITlabTextAlignment
  • HTR+ can be trained further - also with changed CharMap
  • bugfix load dictionary: now also possible for ";"-seperator with headlines
  • feature-request: logging in multi-process-use possible.
  • newest versions of all citlab and planet libs
  • bugfix fail-save Baseline2Polygon - 2 fallbacks for fails

2.0.2

  • MOVE REPOSITORY WITHOUT HISTORY FROM Transkribs to CITlabRostock AND MAKE OPENSOURCE
  • bugfix: check System.getenv() instead of System.getProperty() for $PYTHONPATH
  • bugfix: do not set train_size_per_epoch instad of "-1"
  • bugfix: add process listener to TrainHTRPlus to make status/process observable
  • feature: language resource can be created when trainingdata are created (set property "create_lr"==>"true" or "path_to/file"

2.0.1

  • bugfix: master-confict

2.0.0

  • add new HTR+ which requires CUDA and Tensorflow
  • switch from file-structure to folder-structure fo HTR
  • switch from 1 big planet jar to planet artifacts

1.1.9

  • feature request: make Filename of ConfMatContainer configurable via properties

1.1.8

  • bugfix erronious calculation of coords for very small baselines
  • patch ignore bidi control characters
  • image type: indexed_byte is supported
  • bugfix LA module (array out of bounds error)
  • memory leak fixed?!
  • fix concerning erroneous baselines
  • reduced number of points for short baselines

1.1.6

  • bugfix erroneous split of baseline in LA module

1.1.5

  • adapted the scaling prior to LA module
  • BaseLine2Polygon uses angle of baseline, not angle of region for calculation
  • character \n are ignored when getting TextLine->TextEquiv->Unicode
  • minor bugfixes

1.1.4

  • bugfix #27: switch from CenterOfMass to Average
  • better logging and statistic for Text2Image

1.1.3

  • bugfix #27: line are sorted correctly

1.1.2

  • make kws-group (?ABC) optional

1.1.1

  • bugfix charmap for advanced ATR can be set, if charmap is equal

1.1

  • make dependent on TranskribusErrorRate 2.2.3

  • enhance KWS-quality

  • update advanced HTR

  • make KWS useable for large sets

1.0.4

  • improve kWS (calculate maximum of PRE_KW_POST, PRE and POST group)
  • improve GT-creation for KWS

1.0.3

  • KWS improvement:
  • Property "kws_upper" can be set to true
  • Property "kws_expert" can be used to directly set regular expressions
  • Property "kws_part" can be set to find keyword in words
  • additional properties "kws_min_conf", "kws_threads" and "kws_max_anz" can be set

1.0.2

  • bugfix #26: only throw exception when
  • bugfix T2I: do not throw exception if no baseline for matching is found
  • bugfix LA: resources available in jar
  • bugfix B2P: if region too small, nullpointer can be handled

1.0.1

  • add tests for advanced Layout Analysis
  • update dependencies to PageXmlExtractor-0.3
  • delete unsolved dependencies to log4j

1.0.0

  • bugfix KWS is threadsave
  • bugfix HTR: missing baselines are added using BaselineGeneartionHist
  • update GetTrainFile to seperated train- and test-file
  • bugfixes on creating traindata (traindata will be written into subfolders)
  • save jar-version into metadate of PageXml-file
  • bugfix KWS: missing lineID does not result in empty result
  • adding advanced Layout Analysis
  • adding advanced HTR (without training)
  • make module dependent on trensorflow and GCOC

0.3.3

  • Bugfix number of threads in T2I-Train-Workflow
  • delete depricated T2I-Workflow
  • do not allow B2P in T2I-Workflow
  • delete old T2I-methods
  • make LayoutAnalysisParser::process with PcGtsType public

0.3.2

  • switch to java-version 1.8

0.3.1

  • KWS improvements
  • save LineID in ConfmatContainer
  • use TRANSKRIBUS_HTR-properties to generate name for xml metadata

0.3

  • Bugfix Bidi in Text2Image
  • switch to new Interface in KWS
  • Bugfix maxAnz KWS

0.2.2

  • T2I is available with BIDI
  • Hyphenation more is configurable via T2I_Hyphenation-Properties
  • property "train_status" can be set to train only on pages with a specific statuses (e.g."GT;DONE")
  • Semi-Supvervised training adds channels to Network, if channels are not there
  • Accept all images, that are accepted by transkribus 1.3.*

0.2.1

  • bugfix LA: LA works also for binary images
  • making Text2Image available via Interfaces

0.2

  • continue Text2Image
  • add Test-Method de.uro.citlab.module.util.DictionaryTest to test external dictionaries
  • make dependent on stable versions of TranskribusTokenizer and TranskribusErrorRate
  • Integrate TranskribusErrorRate in WER-calculation while training
  • Make HTR Training observable
  • Cleanup of dictionary works

0.1.6

  • cleanup (test-) resources - only corresponding planet_jar-x.x.x.jar is needed
  • feature #22
  • solve bugfix #23

0.1.5

  • solve bugfix #20
  • more accurate handlings of unicode characters

0.1.4

  • starting of using changelog after this version