Release TinyOCR 0.1.0 frozen 3.10 MB OCR models and evidence

BENCHMARK_RESULTS.md

@@ -0,0 +1,122 @@
+# Broader public OCR benchmark comparison
+
+Completed 2026-10-23 (local time; run timestamps are 2026-10-24 UTC). **Verified comparison:** each frozen model completed 5,089 images across SROIE, TextOCR and HierText. The 3.10 MB compressed candidate is 50.957% smaller, with E2E F1 lower by 0.079-0.257 percentage points. Latency is mixed: candidate p50 is 21.23% higher on receipts, 16.16% higher on HierText and 6.91% lower on TextOCR in this single pass. Peak process RAM is effectively unchanged at about 500 MiB. This evaluates frozen pretrained `v6-tiny` and compressed `v6-w8-rec`; no TinyOCR student was trained.
+
+## Coverage and provenance
+
+The comparison uses every entry in three pinned public splits after exclusions defined before inference. These are comparative development results, not a new untouched final test. Original model weights, thresholds, dependency lock and consumed final-test evidence remain unchanged.
+
+| Source / split | Published images | Prior-parent exclusions | Evaluated images | Legible regions | Ignored regions | Pages eligible for CER/WER |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| SROIE mirror / test | 361 | 20 | 341 | 18,418 | 0 | 341 |
+| TextOCR / validation | 3,124 | 97 | 3,027 | 103,502 | 41,040 | 748 |
+| HierText / validation | 1,724 | 3 | 1,721 | 177,889 | 31,936 | 461 |
+
+Total: 5,089 images per model, with 120 prior-parent exclusions. The exclusion set covers all 20 prior SROIE pages, all prior full-scene parents and every possible legacy TextOCR crop parent. Decoded-pixel checks found no additional overlap with the prior evaluation/SRD catalogs. The selected SROIE split contains two exact duplicate image pairs, leaving 339 distinct decoded receipt images; these published entries remain included. HierText and TextOCR share 23 source scene IDs at different decoded resolutions. Scores are reported separately, never pooled.
+
+The source audit retains 495 HierText and 713 TextOCR polygons extending outside the image frame, nine invalid TextOCR polygons, and one zero-area SROIE region. These were found before inference. Coordinates and text were not edited: the official evaluator applies its existing geometry repair, and the zero-area reference remains an unmatched item in the denominator. Source-declared and decoded image dimensions match. Four TextOCR images declare EXIF orientation 1; the rest have no orientation tag. Of 55 source SROIE scans above 25 MP, 53 remain after prior-parent exclusions; the amended 40 MP cap accommodates this stratum without filtering images for size.
+
+Reproducible records: [manifest](eval/benchmarks-v1.json), [source downloads and hashes](results/benchmark-sources.json), [source annotation audit](results/benchmark-source-audit.json), and [TextOCR publisher snapshot](results/benchmark-rights.json). Downloaded images remain local under `data/benchmarks-v1/`.
+
+| Source | Frozen source and publisher terms |
+| --- | --- |
+| SROIE | [jsdnrs mirror at `bffe40c`](https://huggingface.co/datasets/jsdnrs/ICDAR2019-SROIE/tree/bffe40c26759f3376ec2b3ae9031dbba54cd587c); its card declares CC BY 4.0. Mirror corrections and region boxes are retained. |
+| TextOCR | [Official validation release](https://textvqa.org/textocr/dataset/); publisher bundle declares CC BY 4.0. Underlying image rights also apply; image-specific reuse/training review remains separate. |
+| HierText | [Official repository at `70b6620`](https://github.com/google-research-datasets/hiertext/tree/70b6620b2b112597d8219e11eee9773a1403827c); dataset/repository declares CC BY-SA 4.0. Upstream evaluator, license and attribution are preserved in `third_party/hiertext/`. |
+
+## Accuracy
+
+All scores below are percentages; higher is better. PQ multiplies F1 by mean matched-box IoU, incorporating localization tightness. Full precision, recall, counts and scores are in [the machine-readable comparison](results/benchmark-comparison.json).
+
+| Source | Model | Detection F1 | Detection PQ | E2E F1 | E2E PQ |
+| --- | --- | ---: | ---: | ---: | ---: |
+| SROIE regions | `v6-tiny` | 88.584 | 67.118 | 41.036 | 30.104 |
+| SROIE regions | `v6-w8-rec` | 88.598 | 67.123 | 40.779 | 29.911 |
+| TextOCR words | `v6-tiny` | 58.515 | 40.087 | 43.412 | 30.067 |
+| TextOCR words | `v6-w8-rec` | 58.364 | 40.000 | 43.334 | 30.028 |
+| HierText words | `v6-tiny` | 48.392 | 30.827 | 40.330 | 25.745 |
+| HierText words | `v6-w8-rec` | 48.252 | 30.745 | 40.204 | 25.667 |
+
+The candidate loses 0.257 percentage points of E2E F1 on SROIE, 0.079 on TextOCR and 0.127 on HierText. These are small observed losses, not exact accuracy preservation or a statistical equivalence claim. Absolute scene-text performance remains limited.
+
+The unchanged HierText polygon evaluator uses mutual-best polygon matches at IoU >= 0.5. End-to-end (E2E) matches also require exact, case-sensitive text. Predictions overlapping an ignored region over at least half their area are ignored under the evaluator's default rules. SROIE uses native predicted lines against text regions; TextOCR and HierText use RapidOCR's native word boxes. Word boundaries never use reference labels. Six development-fixture/model checks established that requesting words leaves line text and polygons unchanged.
+
+Only HierText uses its native word evaluator here, on an overlap-filtered validation subset. SROIE and TextOCR scores use a shared HierText protocol and must not be presented as native leaderboard results. SROIE's uppercased reference strings make case differences especially consequential for exact-text E2E scoring. Word-box detection can also change with recognition/alignment even though the detector weights are identical.
+
+Post-run geometry inspection found four baseline and seven candidate zero-area native word boxes, with no invalid line polygons. One zero-area word per model belongs to SROIE, where only lines are scored; three baseline and six candidate word boxes therefore reach word scoring. The official evaluator logs these degeneracies and leaves them as unmatched predictions. None were deleted or regenerated. A regression test added after observing the warnings confirms their precision penalty without changing scorer code or predictions. Empty page outputs were also retained: baseline/candidate counts are 0/0 SROIE, 160/172 TextOCR and 17/18 HierText.
+
+## Supplementary page-order diagnostics
+
+All values are percentages; lower is better. Coverage is restricted to the complete-reference page counts in the coverage table.
+
+| Source | Model | Normalized CER | Normalized WER | Case-sensitive CER | Case-sensitive WER |
+| --- | --- | ---: | ---: | ---: | ---: |
+| SROIE regions | `v6-tiny` | 12.452 | 33.115 | 33.944 | 56.407 |
+| SROIE regions | `v6-w8-rec` | 12.549 | 33.660 | 34.008 | 56.684 |
+| TextOCR words | `v6-tiny` | 30.628 | 52.930 | 31.181 | 53.634 |
+| TextOCR words | `v6-w8-rec` | 30.678 | 52.981 | 31.218 | 53.672 |
+| HierText words | `v6-tiny` | 20.833 | 34.396 | 21.149 | 34.757 |
+| HierText words | `v6-w8-rec` | 20.907 | 34.416 | 21.210 | 34.751 |
+
+Normalized CER increases in all three sources. The largest normalized WER change is on receipts: 33.115% to 33.660%, an increase of 0.545 percentage points.
+
+CER/WER are summed edit counts divided by summed reference character/word counts. Normalized scores use NFC, collapsed whitespace and case folding; punctuation and digits remain. Case-sensitive scores are also retained. Both reference and prediction text use the existing geometric row-order heuristic. These diagnostics include reading-order errors and are not canonical scene-text recognition metrics.
+
+Pages containing any ignored region are excluded from these diagnostics: 0 SROIE, 2,279 TextOCR and 1,260 HierText pages. Their detection/E2E scores remain included. The diagnostic subset therefore differs substantially from the full scene benchmark and may favor easier pages. Non-English references remain included; the recognizer is English-first.
+
+## CPU latency and process memory
+
+| Source | Baseline p50 / p95 (ms) | Candidate p50 / p95 (ms) | p50 change | p95 change |
+| --- | ---: | ---: | ---: | ---: |
+| SROIE regions | 217.35 / 346.84 | 263.50 / 436.40 | +21.23% | +25.82% |
+| TextOCR words | 109.45 / 303.65 | 101.89 / 254.12 | -6.91% | -16.31% |
+| HierText words | 166.23 / 383.70 | 193.09 / 495.64 | +16.16% | +29.17% |
+
+The candidate is slower on receipts and HierText in this pass, and faster on TextOCR. This run supplies no general speedup claim. The large per-source timing differences warrant a repeated paired benchmark with the same word-output path before drawing a deployment latency conclusion; no extra runs were used to select more favorable timings.
+
+| Resource | Baseline | Candidate |
+| --- | ---: | ---: |
+| Model artifact bytes | 6,319,431 | 3,099,213 |
+| Model size (decimal MB) | 6.32 | 3.10 |
+| Initialization (ms) | 119.08 | 206.90 |
+| Peak process working set (MiB) | 498.84 | 500.10 |
+| Complete uninterrupted memory coverage | Yes | Yes |
+
+Model artifacts shrink by 50.957%; process RAM does not decrease. Both models complete in distinct, sequential processes: baseline PID 62236, 2026-10-24 03:02:11-03:16:14 UTC; candidate PID 47168, 03:16:35-03:31:22 UTC. Neither run resumed from an earlier segment.
+
+Hardware: AMD Ryzen 9 9900X, Windows, Python 3.12.9. Both ORT sessions and OpenCV use one thread, CPUExecutionProvider, detector maximum side 960, recognition batch size one and no orientation classifier. Each model runs in its own sequential process with two warmups on an existing development receipt and one measured call per new image. Timing includes file decode, OCR, geometric row ordering and native word-box postprocessing; model initialization, download, scoring and journal writes are outside the timed call. p95 is the nearest-rank 95th percentile across images.
+
+Both processes recorded identical package versions: RapidOCR 3.10.0, ONNX Runtime 1.31.0, NumPy 2.5.4, Pillow 12.3.0, Shapely 2.2.0 and absl-py 2.3.1. The original lock and separate benchmark dependency overlay are hash-pinned in both start records.
+
+Peak working set covers the entire model process, including imports, manifest, runtime, decoded images, models and output adaptation. It is not model-only RAM. This is a single desktop pass, not repeated latency certification or phone evidence. The candidate stores int8 recognizer weights but executes FP32 math; reduced file size does not imply integer-compute acceleration or proportional RAM savings.
+
+## Verification evidence
+
+The immutable [pre-inference protocol](BENCHMARK_VERIFICATION.md) retains its declaration-time pending cells; outcomes belong here so its run-pinned hash stays unchanged. Commands run from `C:\Users\reese\tinyocr`. Below, `BENCH` means `uv run --locked --with-requirements benchmark-requirements.txt`; optional benchmark dependencies are isolated from `uv.lock`.
+
+| Obligation | Executed command | Outcome |
+| --- | --- | --- |
+| Full source reconstruction, hashes, exclusions and pixels | `BENCH python -m tinyocr.prepare_benchmark --verify-only` | Passed, exit 0; 5,089 selected / 120 excluded / 23 shared scene IDs |
+| Source geometry and annotation audit | `BENCH python -m tinyocr.audit_benchmark` | Passed, exit 0; saved diagnostics reproduced without label changes or OCR |
+| Scoring, ignored regions, geometry, journals and archive safety | `BENCH python -m unittest discover -s tests_benchmark -v` | Passed, 14 tests, zero skipped; includes observed degenerate-prediction handling |
+| Word adapter preserves line output | `BENCH python -m tinyocr.benchmark --check-adapter` | Passed, six fixture/model pairs; no new benchmark images used |
+| Baseline predictions | `BENCH python -m tinyocr.benchmark --model v6-tiny` | Passed, 5,089 first predictions in one process |
+| Candidate predictions | `BENCH python -m tinyocr.benchmark --model v6-w8-rec` | Passed, 5,089 first predictions in one process |
+| Completed runs do not infer again | Repeat both completed `--model` commands above | Passed, both report completed journals verified with no inference |
+| Official scoring | `BENCH python -m tinyocr.benchmark --score` | Passed, exit 0; all 10,178 predictions scored |
+| Fresh-process full recomputation and integrity checks | `BENCH python -m tinyocr.benchmark --verify-only` | Passed, exit 0; exact stored comparison reproduced without inference |
+| Existing unit suite | `uv run --locked python -m unittest discover -s tests -v` | Passed, 53 tests, zero skipped |
+| Existing SRD intake | `uv run --locked python -m tinyocr.intake --verify-only` | Passed, 200 candidate entries / 198 unique images; benchmark approval remains false |
+| Existing robustness evidence | `uv run --locked python -m tinyocr.evaluate_robustness --verify-only` | Passed, 80 items per model; original failures retained |
+| Consumed final evidence | `uv run --locked python -m tinyocr.evaluate_final --verify-only` | Passed, 48 items per model; frozen transfer check unchanged |
+| Patch whitespace | `git diff --check` | Passed, exit 0; only existing Windows line-ending notices |
+
+Documentation links, completion markers, new-code syntax and whitespace checks passed. A path-scoped `git diff --exit-code` confirmed that `uv.lock`, `pyproject.toml`, baseline/common/metrics code, the original final protocol/results and candidate lock remain unchanged.
+
+Manual inspection covered publisher snapshots and evaluator matching/ignore/geometry code, the first stored source image and reference regions in each dataset, the six adapter records, and representative saved line/word coordinates. It does not replace independent human label adjudication. Empty detections remain in prediction journals and scores.
+
+Supplementary assertions cross-checked every scored ground-truth total and complete-reference page count against the independently saved source audit, confirmed F1 = 2TP/(ground-truth + prediction count), and checked that process IDs differ, run time windows do not overlap, and both completion records have zero resumed items. All passed. Evaluator geometry warnings described above are retained and covered by explicit scoring tests; they are not skipped examples or failed commands.
+
+First predictions, start fingerprints and process completion records are in `runs/benchmarks-v1/`. A completed `--model` invocation verifies saved evidence without running inference. Do not delete journals or casually alter fingerprinted code/protocol to rerun this experiment. Recompute scores with `--verify-only`; the machine-readable comparison is [results/benchmark-comparison.json](results/benchmark-comparison.json).
+
+Public-pretraining overlap, near-duplicate source lineage beyond the recorded checks, independent real capture coverage and fully human-adjudicated fresh labels remain open. These measurements do not establish readiness for release or create a fresh final partition. The separate collection backlog remains in [.claude/TODO.md](.claude/TODO.md).

LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

NOTICE

@@ -0,0 +1,29 @@
+TinyOCR
+Copyright 2026 Maureese Williams
+
+This product includes PaddleOCR / PP-OCRv6 model artifacts.
+Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
+https://github.com/PaddlePaddle/PaddleOCR
+Licensed under the Apache License, Version 2.0 (see LICENSE).
+
+PP-OCRv6_det_tiny.onnx is the unchanged pretrained detector.
+PP-OCRv6_rec_tiny.onnx WAS MODIFIED by Maureese Williams in 2026:
+selected float32 weight tensors were stored as per-channel int8 values with
+constant decoding back to float32. Computation remains float32. No training
+or calibration was performed. This modified file is not an official
+PaddlePaddle release. SHA-256 hashes and origins are in assets/models.json.
+
+Original ONNX artifacts were distributed through RapidAI/RapidOCR v3.10.0:
+https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv6/det/PP-OCRv6_det_tiny.onnx
+https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv6/rec/PP-OCRv6_rec_tiny.onnx
+Official Apache-2.0 model cards:
+https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_det_onnx
+https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_rec_onnx
+
+Runtime inference uses the separately installed RapidOCR project (Apache-2.0):
+Copyright RapidAI contributors. https://github.com/RapidAI/RapidOCR
+Other installed dependencies retain their own licenses.
+
+Research repository only: the vendored Google HierText evaluator has its own
+CC BY-SA 4.0 license and notices in third_party/hiertext. It and evaluation
+datasets are not included in the TinyOCR runtime distribution.

PP-OCRv6_det_tiny.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f42c0fbd294d95eac1a550e131b277dac97462c8025fa4b6c3cec1b7894bd3d5
+size 1829618

PP-OCRv6_rec_tiny.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38cd559cb4b24c5056592d981b13e086f9472db9b3c8968296f0d0c2b4a65531
+size 1269595

README.md

@@ -0,0 +1,116 @@
+---
+license: apache-2.0
+language:
+  - en
+pipeline_tag: image-to-text
+tags:
+  - ocr
+  - onnx
+  - paddleocr
+  - rapidocr
+  - cpu
+  - quantized
+base_model:
+  - PaddlePaddle/PP-OCRv6_tiny_det_onnx
+  - PaddlePaddle/PP-OCRv6_tiny_rec_onnx
+base_model_relation: quantized
+datasets:
+  - jsdnrs/ICDAR2019-SROIE
+model-index:
+  - name: TinyOCR v6-w8-rec
+    results: []
+---
+
+# TinyOCR 0.1.0
+
+English-first, offline CPU OCR with **3.10 MB of ONNX model files**. This is a
+compressed pretrained PaddleOCR PP-OCRv6 model pair, not a newly trained TinyOCR
+student. The detector is unchanged; the recognizer stores selected weights as
+per-channel int8 values with constant decoding to float32. Computation remains
+float32. No training or calibration was performed. The dataset tag describes
+evaluation, not TinyOCR training.
+
+## Use
+
+Python 3.12:
+
+```sh
+pip install tinyocr-onnx==0.1.0
+tinyocr receipt.png
+tinyocr receipt.png --json
+```
+
+```python
+from tinyocr import TinyOCR
+
+ocr = TinyOCR()
+result = ocr("receipt.png")
+print(result.text)
+print(result.lines)
+print(result.words)
+```
+
+The wheel bundles these exact model files and verifies their SHA-256 hashes.
+It needs no model download or HF account after installation. Dependencies and
+runtime memory are larger than the model files. Inputs are local image paths;
+outputs contain text, line polygons/xyxy bounds and native word polygons in
+original-image pixels. See the [runtime documentation](https://github.com/MMWilliams/tinyocr/tree/master/python).
+
+## Artifacts and provenance
+
+| File | Bytes | SHA-256 |
+| --- | ---: | --- |
+| `PP-OCRv6_det_tiny.onnx` | 1,829,618 | `f42c0fbd294d95eac1a550e131b277dac97462c8025fa4b6c3cec1b7894bd3d5` |
+| `PP-OCRv6_rec_tiny.onnx` | 1,269,595 | `38cd559cb4b24c5056592d981b13e086f9472db9b3c8968296f0d0c2b4a65531` |
+
+Total: 3,099,213 bytes, versus 6,319,431 original bytes (50.957% smaller).
+Original recognizer SHA-256: `e16e242de5937ad92609223f19bc2aff3727ee40b095f996907c24749bad251b`.
+Source ONNX files came from RapidAI/RapidOCR v3.10.0 on ModelScope. NOTICE gives
+the exact URLs and upstream attribution. `export-provenance.json` records
+modified tensors; the export code is in the
+[source repository](https://github.com/MMWilliams/tinyocr).
+
+## Measured results
+
+5,089 images per model with 120 prior-parent exclusions. These are comparative
+development results; public pretraining overlap is unknown.
+
+| Dataset | Images | Original E2E F1 (%) | TinyOCR E2E F1 (%) | Candidate p50 change |
+| --- | ---: | ---: | ---: | ---: |
+| SROIE | 341 | 41.036 | 40.779 | +21.23% |
+| TextOCR | 3,027 | 43.412 | 43.334 | -6.91% |
+| HierText | 1,721 | 40.330 | 40.204 | +16.16% |
+
+Small observed E2E F1 losses of 0.079–0.257 percentage points, not exact accuracy
+preservation. SROIE/TextOCR use the shared HierText polygon evaluator, not their
+native leaderboard protocols. HierText uses its native word evaluator on the
+overlap-filtered subset. Latency is one pass per model on the same Windows CPU
+environment, not a repeated speed study. Peak process working set was
+498.84 MiB original / 500.10 MiB candidate. **No general speedup or RAM saving is
+claimed.** See the [full benchmark](https://github.com/MMWilliams/tinyocr/blob/master/BENCHMARK_RESULTS.md)
+for counts, source rights, protocols, timing and caveats. Images/labels are not
+included in this model repository.
+
+## Configuration and limitations
+
+RapidOCR 3.10.0; ONNX Runtime 1.31.0; one-thread CPU sessions and OpenCV; detector
+ImageNet normalization, maximum side 960, threshold 0.2, box threshold 0.4,
+unclip 1.4, no dilation; recognizer batch one; no orientation classifier.
+The installed API matches the frozen research text/line/word adapter on three
+development fixtures. English-first evaluation does not establish multilingual
+performance. Rotations, difficult scenes and complex layouts can fail.
+Reading order is geometric, not layout understanding. Native word polygons can
+have zero area; they are retained for benchmark parity. No mobile performance
+claim is made. A consumed final comparison is preserved in the research repo;
+future tuning needs a fresh, independently labeled final partition.
+
+## License and links
+
+Apache-2.0 for the runtime and PaddleOCR-derived models. See LICENSE and NOTICE
+for PaddlePaddle / RapidAI attribution and the recognizer modification notice.
+No affiliation with or endorsement by PaddlePaddle or RapidAI is implied.
+
+- [GitHub source and full evaluation reports](https://github.com/MMWilliams/tinyocr)
+- [PyPI tinyocr-onnx](https://pypi.org/project/tinyocr-onnx/)
+- [Original detector card](https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_det_onnx)
+- [Original recognizer card](https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_rec_onnx)

export-provenance.json

@@ -0,0 +1,491 @@
+{
+  "model": "v6-w8-rec",
+  "onnx_version": "1.23.2",
+  "mode": "w8-rec",
+  "method": "original float32 detector + int8 per-channel recognizer",
+  "compute": "float32; constant weight decoding can fold at load; no integer kernel or RAM claim",
+  "calibration": "none",
+  "training": "none",
+  "artifacts": [
+    {
+      "path": "models/v6-w8-rec/PP-OCRv6_det_tiny.onnx",
+      "bytes": 1829618,
+      "sha256": "f42c0fbd294d95eac1a550e131b277dac97462c8025fa4b6c3cec1b7894bd3d5",
+      "source_sha256": "f42c0fbd294d95eac1a550e131b277dac97462c8025fa4b6c3cec1b7894bd3d5",
+      "weights": []
+    },
+    {
+      "path": "models/v6-w8-rec/PP-OCRv6_rec_tiny.onnx",
+      "bytes": 1269595,
+      "sha256": "38cd559cb4b24c5056592d981b13e086f9472db9b3c8968296f0d0c2b4a65531",
+      "source_sha256": "e16e242de5937ad92609223f19bc2aff3727ee40b095f996907c24749bad251b",
+      "weights": [
+        {
+          "name": "conv2d_0.w_0",
+          "shape": [
+            24,
+            3,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 648,
+          "max_abs_error": 0.007004380226135254
+        },
+        {
+          "name": "conv2d_1.w_0",
+          "shape": [
+            48,
+            24,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 10368,
+          "max_abs_error": 0.006679173558950424
+        },
+        {
+          "name": "conv2d_43.w_0",
+          "shape": [
+            48,
+            1,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 432,
+          "max_abs_error": 0.03964325413107872
+        },
+        {
+          "name": "conv2d_4.w_0",
+          "shape": [
+            12,
+            48,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 576,
+          "max_abs_error": 0.0086698979139328
+        },
+        {
+          "name": "conv2d_5.w_0",
+          "shape": [
+            48,
+            12,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 576,
+          "max_abs_error": 0.006149396300315857
+        },
+        {
+          "name": "conv2d_44.w_0",
+          "shape": [
+            96,
+            48,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 4608,
+          "max_abs_error": 0.4647179841995239
+        },
+        {
+          "name": "conv2d_45.w_0",
+          "shape": [
+            48,
+            96,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 4608,
+          "max_abs_error": 0.0030015218071639538
+        },
+        {
+          "name": "conv2d_47.w_0",
+          "shape": [
+            48,
+            1,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 432,
+          "max_abs_error": 0.016505658626556396
+        },
+        {
+          "name": "conv2d_48.w_0",
+          "shape": [
+            96,
+            48,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 4608,
+          "max_abs_error": 0.02811253070831299
+        },
+        {
+          "name": "conv2d_49.w_0",
+          "shape": [
+            48,
+            96,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 4608,
+          "max_abs_error": 0.01393282413482666
+        },
+        {
+          "name": "conv2d_50.w_0",
+          "shape": [
+            48,
+            1,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 432,
+          "max_abs_error": 0.006548999343067408
+        },
+        {
+          "name": "conv2d_51.w_0",
+          "shape": [
+            96,
+            48,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 4608,
+          "max_abs_error": 0.010334327816963196
+        },
+        {
+          "name": "conv2d_52.w_0",
+          "shape": [
+            96,
+            96,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 9216,
+          "max_abs_error": 0.0068708136677742004
+        },
+        {
+          "name": "conv2d_54.w_0",
+          "shape": [
+            96,
+            1,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 864,
+          "max_abs_error": 0.026382118463516235
+        },
+        {
+          "name": "conv2d_17.w_0",
+          "shape": [
+            24,
+            96,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 2304,
+          "max_abs_error": 0.007207766175270081
+        },
+        {
+          "name": "conv2d_18.w_0",
+          "shape": [
+            96,
+            24,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 2304,
+          "max_abs_error": 0.005261123180389404
+        },
+        {
+          "name": "conv2d_55.w_0",
+          "shape": [
+            192,
+            96,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 18432,
+          "max_abs_error": 0.03450369834899902
+        },
+        {
+          "name": "conv2d_56.w_0",
+          "shape": [
+            96,
+            192,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 18432,
+          "max_abs_error": 0.009590744972229004
+        },
+        {
+          "name": "conv2d_58.w_0",
+          "shape": [
+            96,
+            1,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 864,
+          "max_abs_error": 0.005693256855010986
+        },
+        {
+          "name": "conv2d_59.w_0",
+          "shape": [
+            192,
+            96,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 18432,
+          "max_abs_error": 0.014728039503097534
+        },
+        {
+          "name": "conv2d_60.w_0",
+          "shape": [
+            96,
+            192,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 18432,
+          "max_abs_error": 0.02412763237953186
+        },
+        {
+          "name": "conv2d_61.w_0",
+          "shape": [
+            96,
+            1,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 864,
+          "max_abs_error": 0.008419990539550781
+        },
+        {
+          "name": "conv2d_62.w_0",
+          "shape": [
+            192,
+            96,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 18432,
+          "max_abs_error": 0.006885454058647156
+        },
+        {
+          "name": "conv2d_63.w_0",
+          "shape": [
+            160,
+            192,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 30720,
+          "max_abs_error": 0.0031155943870544434
+        },
+        {
+          "name": "conv2d_65.w_0",
+          "shape": [
+            160,
+            1,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 1440,
+          "max_abs_error": 0.15238910913467407
+        },
+        {
+          "name": "conv2d_30.w_0",
+          "shape": [
+            40,
+            160,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 6400,
+          "max_abs_error": 0.008130047470331192
+        },
+        {
+          "name": "conv2d_31.w_0",
+          "shape": [
+            160,
+            40,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 6400,
+          "max_abs_error": 0.012635335326194763
+        },
+        {
+          "name": "conv2d_66.w_0",
+          "shape": [
+            320,
+            160,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 51200,
+          "max_abs_error": 0.030289411544799805
+        },
+        {
+          "name": "conv2d_67.w_0",
+          "shape": [
+            160,
+            320,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 51200,
+          "max_abs_error": 0.006284937262535095
+        },
+        {
+          "name": "conv2d_69.w_0",
+          "shape": [
+            160,
+            1,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 1440,
+          "max_abs_error": 0.0024935081601142883
+        },
+        {
+          "name": "conv2d_70.w_0",
+          "shape": [
+            320,
+            160,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 51200,
+          "max_abs_error": 0.009998321533203125
+        },
+        {
+          "name": "conv2d_71.w_0",
+          "shape": [
+            160,
+            320,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 51200,
+          "max_abs_error": 0.009426265954971313
+        },
+        {
+          "name": "conv2d_73.w_0",
+          "shape": [
+            160,
+            1,
+            3,
+            3
+          ],
+          "axis": 0,
+          "elements": 1440,
+          "max_abs_error": 0.03488779067993164
+        },
+        {
+          "name": "conv2d_74.w_0",
+          "shape": [
+            320,
+            160,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 51200,
+          "max_abs_error": 0.010606534779071808
+        },
+        {
+          "name": "conv2d_75.w_0",
+          "shape": [
+            160,
+            320,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 51200,
+          "max_abs_error": 0.12106248736381531
+        },
+        {
+          "name": "p2o.pd_op.unsqueeze.0.0",
+          "shape": [
+            160,
+            1,
+            1,
+            5
+          ],
+          "axis": 0,
+          "elements": 800,
+          "max_abs_error": 0.0030310668516904116
+        },
+        {
+          "name": "p2o.pd_op.unsqueeze.2.0",
+          "shape": [
+            160,
+            160,
+            1,
+            1
+          ],
+          "axis": 0,
+          "elements": 25600,
+          "max_abs_error": 0.005150206387042999
+        },
+        {
+          "name": "linear_0.w_0",
+          "shape": [
+            160,
+            80
+          ],
+          "axis": 1,
+          "elements": 12800,
+          "max_abs_error": 0.009965181350708008
+        },
+        {
+          "name": "linear_1.w_0",
+          "shape": [
+            80,
+            6906
+          ],
+          "axis": 1,
+          "elements": 552480,
+          "max_abs_error": 0.003063395619392395
+        }
+      ]
+    }
+  ],
+  "artifact_bytes": 3099213
+}

models.json

@@ -0,0 +1,15 @@
+{
+  "model": "v6-w8-rec",
+  "artifacts": [
+    {
+      "filename": "PP-OCRv6_det_tiny.onnx",
+      "bytes": 1829618,
+      "sha256": "f42c0fbd294d95eac1a550e131b277dac97462c8025fa4b6c3cec1b7894bd3d5"
+    },
+    {
+      "filename": "PP-OCRv6_rec_tiny.onnx",
+      "bytes": 1269595,
+      "sha256": "38cd559cb4b24c5056592d981b13e086f9472db9b3c8968296f0d0c2b4a65531"
+    }
+  ]
+}