american-stories-onnx

American Stories — ONNX layout/line models

ONNX detectors for historical-newspaper layout analysis, used by newspaper-ocr as the AsYoloDetector backend. Mirrored here so the detector can download them on demand instead of depending on a local Dropbox path.

From the American Stories project (dell-research-harvard/AmericanStories, NeurIPS 2023). YOLOv8 models trained on Library of Congress Chronicling America scans; upstream distributes them via a Dropbox folder. Public-domain source.

Files

• layout_model_new.onnx — page layout/region detector (~104 MB, md5 21d242c6…).
• line_model_new.onnx — line detector (~44 MB, md5 23fccd7b…). Re-fetched from the official Dropbox after an earlier local copy had been overwritten with the layout model.

Usage

from newspaper_ocr.detectors.as_yolo import AsYoloDetector
det = AsYoloDetector()   # auto-downloads from this repo if not cached locally