Releases: ahmedosaka/kotonoha
Release list
OCR
Book text OCR model v1
Detector, recognizer and character dictionary for the book OCR engine — the
models that read scanned Japanese books.
Play installs receive these as the on-demand asset pack book_ocr_pack. This
release is the fallback route: sideloaded builds, and any install where Play
cannot deliver the pack, fetch these files instead. The app verifies every file
against the hashes below before use, so a mirror serving different bytes fails
rather than quietly substituting a different model.
| file | sha256 | bytes |
|---|---|---|
| book_det.onnx | d9056b16f446cd313875244fd37c935fc94f05c68946e3bc85a8a2c4db5ac068 |
4,766,440 |
| book_rec.onnx | d7c4086072175f7d24e2b0185ca4fc79623d5a612421c25908e358b0d263b54f |
16,536,382 |
| book_dict.txt | d1979e9f794c464c0d2e0b70a7fe14dd978e9dc644c0e71f14158cdf8342af1b |
74,012 |
Total 21,376,834 bytes.
Provenance
Produced by tool/export_book_ocr_models.py from the upstream project's own
Apache-2.0 release at pinned revisions, converted locally with a pinned
toolchain (paddlepaddle 3.0.0, paddle2onnx 2.0.1, onnx 1.17.0, opset 16). Two
clean-room runs of that script produce these exact bytes. The character
dictionary is byte-identical to the upstream file.
Upstream is PP-OCRv5 mobile, published by PaddlePaddle under the Apache License
2.0. Attribution ships in the app at assets/licenses/NOTICE-book-ocr.txt and
on its open-source licences screen.
Book text OCR model v1 (prerelease test)
PRERELEASE TEST ARTIFACT — not a production release.
Detector/recognizer/dictionary for the book OCR engine, produced by
tool/export_book_ocr_models.py from upstream Apache-2.0 sources at pinned
revisions. Two clean-room exports produce these exact bytes.
book_det.onnx d9056b16f446cd313875244fd37c935fc94f05c68946e3bc85a8a2c4db5ac068 4766440
book_rec.onnx d7c4086072175f7d24e2b0185ca4fc79623d5a612421c25908e358b0d263b54f 16536382
book_dict.txt d1979e9f794c464c0d2e0b70a7fe14dd978e9dc644c0e71f14158cdf8342af1b 74012
Licence and attribution: assets/licenses/NOTICE-book-ocr.txt in the app.