OCR-D/ocrd_all v2023-06-14

on GitHub

Changed:

• All docker images now contain git checkouts and retain /build, i.e. behave like the -git variants
• No more git updates within docker build, but fix git module dependency outside
• Reduce docker image size (by reinstating all-in-one layer, removing cache, avoiding duplicate CUDA libraries...)
• Use git submodule update --single-branch on CI to reduce docker image size

Added:

• make deps-cuda: non-intrusively support CUDA system dependencies (in docker or native)
• make ocrd-all-tool.json: Generate and upload a combination of all processors' ocrd-tool.json, #362
• make test-workflow: Run a workflow with most processors as a general smoke test
• make test-cuda: to test whether CUDA properly set up and has GPU available
• make test-core: Run OCR-D/core unit tests

Fixed:

• dependencies between modules, esp. with custom OCRD_MODULES selection
• editable mode (pip install -e)
• OpenCV build
• get tesserocr from PyPI if disabled as a module
• get ocrd from PyPI if core disabled as a module
• consistent interoperable module versions (esp. Numpy/OpenCV/Shapely/Protobuf/Torch/TF Python dependencies)

cor-asv-ann 006a70e..2c4b1ff

Release: v0.1.14

• CI: use ocrd/core-cuda as base image
• CI: dummy venv
• CI: use proper tab character
• CI: clone first
• CI: mkdir first
• CI: chdir to tmp location
• CI: use /tmp for aux clone of ocrd_all
• try getting tensorflow-gpu from Nvidia
• use proper URLs for submodules
• Merge pull request #6 from kba/init-report-dict
• evaluate: skip pages with no results

core de08453..6708624

Release: v2.51.0

• Merge pull request #1055 from bertsky/deps-cuda
• ci: disable upterm for gh actions
• readme: remove dockerhub/travis badge, add GH actions badge
• debug gh actions
• test bashlib: /usr/bin/env bash instead of /bin/bash
• test_workspace_bagger: use ocr-d.de instead of google.com for testing
• disable logging tests until properly fixed
• docker-image: reuse local ghcr.io image instead of docker.io
• 📦 v2.51.0
• 📝 changelog
• make help: improve description
• Revert "Merge remote-tracking branch 'hnesk/no-more-pkg_resources' into release-2.36.0"
• remove out-dated processor resources
• docker-cuda: improve (reduce size) again…
• docker-cuda: rewrite…
• core-cuda: use same CUDA libs as needed for Torch anyway
• Merge branch 'pr-1008' into reduce-cuda
• Merge branch 'master' of https://github.com/OCR-D/core into reduce-cuda
• make install on py36: revert to prefer-binary via install
• make install on py36: fix prefer-binary syntax
• make install on py36: prefer binary OpenCV/Numpy via pip config instead of preinstall
• core-cuda: install more CUDA libs via pip and ld.so.conf, simplify Dockerfile for that
• core-cuda: use CUDA 11.8, install cuDNN via pip and make available system-wide via ld.so.conf
• reinstate workaround for shapely, but more robust
• docker-cuda: change base image, no multi-CUDA runtimes
• keep gcc, no autoremove
• rehash after pip upgrade
• give up workaround for shapely-CUDA issue

dinglehopper 0fd4ea1..35be58c

• Merge pull request #83 from INL/feat/batch-processing
• Merge pull request #82 from CircleCI-config-suggestions-bot/StoreTestResults
• 🧹 .gitignore .python-version (for pyenv)
• 🧹 Remove qurator. namespace prefix
• 🐛 Fix installing by calling find_namespace_packages in setup.py
• 🕸Do not use deprecated ID, pageId options
• 🔧 Remove explicit namespace_packages
• ✔ CircleCI: Explicitly install binary opencv-python-headless (dep of OCR-D?) to avoid compilation
• 🐛 Remove deprecated declare_namespace call

eynollah ea792d1..706433c

Release: v0.2.0

• Revert "Merge pull request #97 from qurator-spk/420-namespace-package"
• Merge pull request #100 from bertsky/patch-2
• Merge pull request #97 from qurator-spk/420-namespace-package

ocrd_cis c90b29f..a0ea0a2

Release: v0.1.5

• Merge branch 'kba:typo' #91 into fix-alpha-shape
• Merge branch 'kba:double-page-max-size' #96 into fix-alpha-shape
• Merge branch 'kba:resolve-resources' #83 into fix-alpha-shape
• segment: adapt to OpenCV changes
• resegment (baseline/ccomps): improve handling of fg conflicts
• resegment: add param baseline_only
• check_page/region/line: skip assumptions on number of components
• adapt to Shapely 2.0 deprecations
• adapt to Numpy 1.24 dtypes
• resegment: list instead of generator
• re/segment: improve polygon simplification
• re/segment: join_baselines: skip lines outside of polygon
• re/segment: join_baselines: for complex subtypes, apply recursively
• re/segment: join_polygons: connect touching neighbours, too

ocrd_fileformat dacfa50..4e7e0de

Release: v0.7.0

• 📦 v0.7.0
• update ocr-fileformat

ocrd_kraken 802c6b0..b13dd8a

Release: v0.3.0

• segment/recognize: default to device=cuda:0 (now backed by safe fall-back)
• segment/recognize: fall back to CPU if no CUDA device
• fix typo
• update changelog
• recognize: project text upwards in order by concatenation
• recognize: ensure baseline/boundary are consistent
• recognize: ignore invalid baselines
• setup metadata: update/improve
• deps-ubuntu: update
• improve/update readme
• Dockerfile: use CUDA base image, improve labels
• update changelog
• recognize: pass lines in baseline format if any baselines are annotated
• update blla.model URL (master→main)
• recognize: workaround for empty/failed line records
• recognize: workaround for better quality box cuts
• recognize: avoid invalid polygons on single-glyph words
• Revert "recognize: avoid invalid polygons on single-glyph words"
• segment: also show tags/type prediction
• recognize: avoid invalid polygons on single-glyph words
• recognize: use proper data structures of rpred

ocrd_pagetopdf 6155605..4f4a330

Release: v1.0.0

• Merge pull request #22 from bertsky/fix-input-files

ocrd_wrap 63c04d5..2cd800d

Release: v0.1.8

• 📦 0.1.8
• Merge pull request #10 from bertsky/update-numpy

opencv-python 6b73d90..474a1cc

Release: 72

• Merge pull request #849 from asmorkalov/as/python3_for_build
• Fix: numpy version for python 3.11 (#839)
• Merge pull request #852 from asmorkalov:as/ci_check
• Merge pull request #837 from bertsky/fix-py38-build
• Merge pull request #838 from henryiii/patch-2

sbb_binarization 39ef3fd..010ec99

Release: v0.1.0

• 📦 v0.1.0
• Update README.md
• update CI badge
• Merge pull request #59 from bertsky/change-model-url
• Merge pull request #56 from bertsky/non-verbose

workflow-configuration cb923f7..5aff777

• ocrd-import: add option --regex (positive path selector)
• ocrd-import: fix skipping in subshell
• add METS transforms to TOC
• generalise standalone CLI for both PAGE and METS XSL, update documentation
• mets-copy-agents.xsl: make path for other-mets relative to input mets (not stylesheet file)
• (ocrd-)page-transform: add pretty-printing option
• add page-ensure-readingorder.xsl
• add page-ensure-textequiv-index.xsl
• ocrd-import: also replace comma in IDs
• Merge remote-tracking branch 'origin/master'
• page-textequiv-*: ensure target TextEquiv exists