sebastien
0a4f4a80d0
Bundle the trained heatmap/image orientation model and run it after provisional blank-page and split analysis. Confirm candidates with Gemini using low thinking; resolve mixed A3 sheet ordering with the complete statement and medium thinking. Keep A4 scans independent and export paired A3 scans in booklet reading order. Reanalyse uniformly inverted source sheets, validate ordering permutations, and preserve source provenance, corrected rotations and reading slots in manual review. Document the experiments, rejected approaches, calibration and runtime limits. Validation: focused splitting/orientation regression suites pass; bundled model checksum and wheel assets verified. Full Interro07 offline run: 44.26 seconds, eight candidates, mocked upright confirmations, source hashes and geometry intact.
709 B
709 B
Handwriting orientation runtime assets
model.joblib contains the trained RBF SVM and ExtraTrees fusion head.
Training assessments: Interro01, 02, 03, 05. Calibration: Interro06.
The conservative operating threshold is 0.3044920634920634.
doc.onnx is PaddlePaddle's PP-LCNet_x1_0_doc_ori, Apache-2.0:
https://huggingface.co/PaddlePaddle/PP-LCNet_x1_0_doc_ori_onnx
settings.json retains thresholds, model hashes and validation provenance.
Do not replace weights without matching preprocessing and recalibration.
The runtime uses scikit-learn 1.9.1 and ONNX Runtime 1.31.1.
See docs/page_orientation.md and docs/orientation_final_training.md for the algorithm, experiments, evidence and limitations.