v0.1.0

Fixed:

• repaired operation_level=region (typo)
• repaired standalone CLI --version (pkg name)
• repaired standalone CLI channel format (uint8)

Changed:

• Trained models loadable and registered in SavedModel format
• Test both models, deployed normally (not in CWD)
• Test input with actual regions in operation_level=region

v0.0.11

Added:

• Trained models listed in ocrd-tool.json for download with OCR-D resource manager, #53

v0.0.10

Fixed:

• Use correct import, s/click/types, #40

v0.0.9

Changed:

• Factor setup by @bertsky in #31
• actually use TF2 (with TF1.compat mode) by @cneud in #35

v0.0.8

Fixed:

• handle image smaller than patch size
• fix unbound variable error, #27

v0.0.7

Changed:

• Use OCR-D/core resource resolving, #25

v0.0.6

Fixed:

• Require h5py < 3, qurator-spk/sbb_textline_detection#50, #18
• Require tensorflow-gpu (CPU+GPU), not tensorflow (CPU only), #20

v0.0.5

Fixed:

• Memory leak, start tf session only once, #17 ht @sulzbals

v0.0.4

Changed:

• Env var SBB_BINARIZE_DATA is combined with model param now, #9

v0.0.3

Fixed:

• typo broke sbb_binarize CLI, #13