Commit Graph

730 Commits (b4df978dd574d5cc950b4319dd7bc568b825df83)
 

Author SHA1 Message Date
Robert Sachunsky 335aa273a1 simplify, wrap extremely long lines
Robert Sachunsky cfc65128b1 reduce redundancy/indentation
Robert Sachunsky 01376af905 do_order_of_regions_with_model: simplify
vahidrezanezhad 92bfac4b41 Provide OCR as an option to process a directory of XML files, incorporating layout and text line coordinates.
vahidrezanezhad fbeef79d50 adding scatter_nd inference
Robert Sachunsky 0ae28f7d3e switch from stdlib to loky.ProcessPoolExecutor, ensure shutdown
vahidrezanezhad f93c6c288d function of patch-wise inference with scatter_nd is added
vahidrezanezhad 0e8c561618 debugging issues
Robert Sachunsky e9c0d716f6 CI: install optional dependencies, too
Robert Sachunsky dcaf796283 change polarity of orientation angle (PAGE schema required cw=positive)
Robert Sachunsky b4b0890294 add option to overwrite output xml, but skip by default if file exists
Robert Sachunsky b9ca7a6191 log num_cols-dependent resizing
Robert Sachunsky 9270ea4550 annotate region angles in PAGE
Robert Sachunsky 3b70b11ea6 avoid deskewing patches if binary-empty
Robert Sachunsky 7e9ee90e6e switch from (ad-hoc) mp.Pool to (attribute) concurrent.futures.ProcessPoolExecutor
Robert Sachunsky 68456ea002 do_work_of_slopes_new*, do_back_rotation_and_get_cnt_back, do_work_of_contours_in_image: use mp.Pool, simplify
Robert Sachunsky 25e967397d exit early if no text regions found (to avoid segfault)
Robert Sachunsky 21efea8711 no del on function argument
Robert Sachunsky 5e0c1da711 simplify
Robert Sachunsky 54cb15056b do_image_rotation / return_deskew_slop: avoid code duplication, simplify via mp.Pool
Robert Sachunsky 6fe02df973 do_image_rotation: fix f93fa12 (do return results)
Robert Sachunsky d68017037c do_prediction: trigger GC to avoid CUDA OOM
Robert Sachunsky ad748d0039 do_prediction: avoid code duplication
Robert Sachunsky c3163caefd avoid indentation
Robert Sachunsky 055463d23a avoid indentation
Robert Sachunsky aaea2ef463 simplify
Robert Sachunsky 3d88b207fc run: log instead of print
Robert Sachunsky a520bd1f77 wrap extremely long lines
Robert Sachunsky cd4e426977 avoid indentation (skip_layout_and_reading_order)
Robert Sachunsky 5b82320707 avoid indentation
Robert Sachunsky 9f12fa241d log-level: only set 'eynollah' logger level
Robert Sachunsky 14beb46224 simplify loading models w/o dir_in mode
Robert Sachunsky 329fac23f6 do not reload enhancement model in dir_in mode, simplify
Robert Sachunsky 3b9a29bc5c simplify dir_in conditionals
Robert Sachunsky 7ae64f3717 RO model: do not reload when in dir_in mode
Robert Sachunsky f765e2603b move Torch to optional dependencies (to avoid clash with TF over CuDNN)
vahidrezanezhad 871d7bfc5a fixed: machine based reading order cause tuple index out of range error if number of textregion is one.
vahidrezanezhad 6aad006f4c filter textregions without textline
kba 1083d1c7fb gha: try to free disk space
vahidrezanezhad 8014a9e416
Update Makefile
vahidrezanezhad 3000255a24
Update Makefile
vahidrezanezhad 1746920275
Update Makefile
vahidrezanezhad b622494f34 new table detection model is integrated
vahidrezanezhad d9f79c3404 fixing IndexError by reading order detection
vahidrezanezhad 5fa8ca46a4 updating requirements
vahidrezanezhad ce5b611296 tests are passed - new models by the way should be uploaded
vahidrezanezhad f43c49c508 textlines of drop capitals are connected to corresponding textline if possible otherwise they are inserted in corresponding textregion
vahidrezanezhad 22b0b07a73 drop capital and marginals extraction is updated
Clemens Neudecker 1ae77e61c8
Update requirements.txt
vahidrezanezhad 8409de0e58 sbb_binarization is integrated into eynollah works in framework of ocrd - sbb_binarization in ocrd works for individual images by the way as standalone flowing from directory can be used now. For eynollah in ocrd framework I have added -light version as default parameter.