v1.0.6

• Fix: Depend on protobuf < 4.0 (@mikegerber, #89)
• Tested on Python 3.6-3.10

v1.0.5

• README: Do not mention obsolete + unsupported checkpoint parameter (@mikegerber, #80)
• Skip tiny or bin-empty lines (@bertsky, #76)

v1.0.4

• 🚨 Breaking change: There is no checkpoint parameter anymore, it was replaced with the new parameter checkpoint_dir (@bertsky, #70)
  • e.g. -P checkpoint "/var/lib/calamari-models/GT4HistOCR/2019-12-11T11_10+0100/*.ckpt.json" becomes -P checkpoint_dir "/var/lib/calamari-models/GT4HistOCR/2019-12-11T11_10+0100/"
• Descend to all available TextRegions recursively (@bertsky, #75)
• This release is now tested on Python 3.6 → 3.10 (@mikegerber)

v1.0.3

• Fix TensorFlow dependency for Python 3.9
• init from constructor not process() (@bertsky)

v1.0.2

• Warn if passing raw images to single-channel models (@bertsky)
• Fix word and glyph coordinates (@mikegerber)

v1.0.1

• ocrd_calamari now supports OCR-D resource resolving, i.e. search paths for models (@kba)
• New parameter checkpoint_dir, to give model directory (not model files) (@kba)
• Bug fix: handle empty line images #48 (@kba)

v1.0.0

• Update to support Calamari 1.0
• Major number increase due to needed model update
• New GT4HistOCR model required
  • https://qurator-data.de/calamari-models/GT4HistOCR/2019-12-11T11_10+0100/
  • This model gives as good results as before (https://gist.github.com/mikegerber/7e16e814bf723f04c47c5a6e2a6ac1d1)
• Calamari 1.0 requires TF 2!

v0.0.7

• Set pcGtsId to fix validation (#40)

v0.0.6

• Fixes for glyph and word segmentation output

v0.0.5

• Fixes for glyph and word segmentation output