astromandarin's picture
Card: crops of real lines with their sources, whole-page reading (read_page, --page), what was taken from han2ids
a9aab5c verified
Raw History Blame Contribute Delete
2.81 kB
"""OpenGlyph-OCR-IDS: read one line of printed scholarly Chinese that holds rare characters, glyphs without a code point
and characters no catalogue has. (MLX, Apple silicon. Run from a local copy of the repository this package lies in.)
python -m openglyph_ocr_ids LINE.png [LINE2.png ...] [--options]
from openglyph_ocr_ids import load, read_line, atoms
ocr = load() # the folder that holds this package; or load('/path/to/the/folder')
out = read_line(ocr, 'line.png')
print(out['text']) # 𤪷也庾參與<gw:u2ff1-u65e5-u82f1><gw:ebag_s093-073><ids>⿰拍嵩</ids>了燒造磚瓦的工作
out = read_line(ocr, 'line.png', options=True)
print(out['options']) # the second option of every place that has one
What comes back (read_line):
text Unicode text; a glyph without a code point that GlyphWiki has is written <gw:NAME> (its GlyphWiki name); a
glyph that has neither is written <ids>IDS</ids>, an Ideographic Description Sequence the model composed
ids every IDS the model wrote and what it spells: a coded character (a rare character read by its parts), a glyph
with a token, or a new glyph
options with options=True: beside the text, the other reading of every place that has one
The picture: one horizontal line of print as cropped from a page (the lines measured are about 100 to 150 pixels high
in the papers, 96 in the special test). A whole page can be read in one go with read_page(ocr, 'page.png') (or
--page): it is then read as OpenGlyph-OCR-0 reads, without IDS; cutting the page into lines reads better.
How a line is read. Four sets of weights work in one model built on Xiaomi-OCR-0: the reader (OpenGlyph-OCR-0, unchanged)
writes the line; the opener says where a glyph needs an IDS; at such a place, and wherever the reader itself writes a rare
character or a glyph token, the glyph is read two more ways -- unfolded (cut into its parts by a program, the parts read
by the part reader) and spelt (by the speller, from the glyph as it stands); what stands is decided by both and by gates
chosen on tuning lines (reading.json). The modules: build (the model, the interface), unfold (the reading of a line),
spell (the speller), unfold_cut and unfold_search (the cuts), canon (canonical forms), decode (constrained decoding).
The versions matter: this was measured with mlx 0.32.3, mlx-vlm 0.3.12, transformers 5.18.0, tokenizers 0.23.2
(requirements.txt); torch and torchvision are there for the pictures only (with torchvision, transformers prepares them
as they were prepared when the model was measured)."""
from .build import load,read_line,read_page,readable,atoms
__all__=['load','read_line','read_page','readable','atoms']