v0.1.8

Fixed

• workarounds for broken models (DocBank_X101, Jambo-sudo_X101)
• make deps: add explicit reqs prior to pip step with Torch index
• set pc:PcGts/@pcGtsId from mets:file/@ID

Added

• CI for CLI tests (with cached models and stored result artifacts)

Changed

• migrated model URLs from external to Github release assets

v0.1.7

Fixed

• adapt to Numpy 1.24 (no np.bool)

Added

• model by Jambo-sudo (PubLayNet+custom GT)
• model by LayoutParser (PRImA Layout GT)
• CLI tests

v0.1.6

Fixed

• avoid colon in generated region IDs
• make deps: add explicit deps for torch
• fix/update file resources
• fix model config base paths on-the-fly

Added

• add Psarpei TD model

v0.1.5

Fixed

• param debug_img: 1 image per page
• URLs/specs for PubLayNet/JPLeoRX models

v0.1.4

Added

• param postprocessing (select steps, including none)
• param debug_img (styles to visualise raw predictions, including none)

v0.1.3

Fixed

• make deps: fall back to Detectron2 src build

Changed

• added various models as file resources
• added corresponding preset files
• updated documentation

v0.1.2

Fixed

• make deps: fix CUDA detection even more
• apply device param as passed

Changed

• downscale images to no more than 150 DPI for prediction (for speed)
• add param operation_level (default page), add table mode

v0.1.1

Fixed

• make deps: fix CUDA detection and allow CPU as fallback

Changed

• instance segmentation postprocessing: use asymmetric overlap
  criterion for non-maximum suppression
• skip instances which belong to classes with empty category
• annotate incrementally (by skipping candidates that overlap
  with pre-existing top-level regions)

v0.1.0

improve introduction