Submission: ShreyShingala - #59
Open
ShreyShingala wants to merge 2 commits into
Open
ShreyShingala wants to merge 2 commits into
ShreyShingala wants to merge 2 commits into
Conversation
Deterministic, fully offline pipeline: RapidOCR (PP-OCRv4 det + PP-OCRv5 rec, ONNX) plus a fine-tuned Tesseract LSTM and a hand-written policy engine. No LLM, VLM, cloud OCR or network at runtime; no hardcoded answers and nothing keyed on case id. 133.65/150 measured inside the submission image under the published scoring contract on the public training set, 11 catastrophic false approvals, 2.32 s/PDF against a 6 s budget. https://claude.ai/code/session_018Jh51xdGJftwD38ScotAqB
ShreyShingala
force-pushed
the
submission-shreyshingala
branch
from
August 3, 2026 23:23
d64e627 to
202491f
Compare
ShreyShingala
force-pushed
the
submission-shreyshingala
branch
from
August 4, 2026 05:51
49d1b59 to
c2abf1d
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
MIB Doc Challenge Submission
Links
Dockerfile): https://github.com/ShreyShingala/ocr-document-pipeline-challengeApproach
Deterministic, fully offline pipeline. No LLM, VLM, cloud OCR or network at runtime; no hardcoded
answers and nothing keyed on case id, filename or hash.
RapidOCR (PP-OCRv4 mobile detector + PP-OCRv5 mobile English recogniser, ONNX Runtime) with a
fine-tuned Tesseract LSTM as a second engine, geometric frame registration that undoes the
generator's band-translation and row-shear damage, closed-vocabulary CTC recovery, and a
hand-written policy engine. The trust boundary sits at parse time: white and near-white ink, text
outside the crop box and hidden layers are dropped while parsing, so injected instructions and
planted answer keys never become candidate evidence.
Result
133.65 / 150 measured inside the submission image under the published scoring contract
(
--cpus 4 --memory 8g --read-only --network none) on the public 1,000-packet training set:70.46 classification, 45.75 extraction, 17.44 calibration. 11 catastrophic false approvals,
0 missing cases, 0 invalid records, 2.32 s/PDF against a 6 s budget.
Out-of-fold (5-fold, refitting the calibration table and the one policy threshold inside each
fold) the total is 133.54, so almost nothing here is in-sample optimism.
Checklist
submissions/ShreyShingala/predictions.jsonl,MEMO.md, andSUBMISSION.mdpredictions.jsonlpassesscripts/validate_submission.pyagainstdata/validation_manifest.csv(5,000 valid records, 0 missing)Dockerfile--network none) and accepts<input_pdf_dir> <output_predictions_path>DOCKER_SUBMISSION.md(largest 14.7 MB, 28 MB total)