Dataset for EMNLP'23 Paper "DocTrack: A Visually-Rich Document Dataset Really Aligned with Human Eye Movement for Machine Reading"
-
Updated
Oct 25, 2023
Dataset for EMNLP'23 Paper "DocTrack: A Visually-Rich Document Dataset Really Aligned with Human Eye Movement for Machine Reading"
Replication package for our eye-tracking study on programmers' linearity of reading order
Post-process PageXMLs to improve their region reading order
Learning to Sort Handwritten Text Lines in Reading Order through Estimated Binary Order Relations
PDF layout intelligence for .NET — structured extraction tuned for RAG and LLM pipelines.
Document Layout Analysis benchmark and comparison suite (58 configurations across 23 model repositories) against your documents in isolated environments, with unified evaluation, consensus, and visual reporting.
Thai document OCR pipeline — BMFL 0.740 on ThaiOCRBench Full-page OCR, above Qwen2.5-VL 72B, no LLM, ~2.7 s/page
OCR in pure Rust: documents, screens, and a change-gated LIVE frame reader. Detection, recognition and reading order computed from geometry, no Python, no C/C++ by default, nothing GPL. Memory safe. The ffai-carmenta crate.
Reports, datasets, and experiments for DeepSolo-inspired Korean OCR reading-order correction.
Derives reading order for multi-column pages, keeping full-width table rows whole and lifting running heads, folios and footnotes out of the flow
Recursive XY-Cut algorithm for spatial reading order and document layout.
A high-quality PDF text extraction library — improving on reading order, font encoding recovery, structured output, and hybrid vector/OCR pipelines
Swift package for Korean Apple Vision OCR reading-order correction with DeepSolo-inspired ordered-point guidance.
An enterprise-grade, high-performance application that parses PDF files from a source directory, extracts their raw text payloads using deep layout analysis (native reading-order reconstruction + multi-column heuristics), and writes corresponding `.txt` and/or `.md` files into a destination directory.
Reading-order scoring for OCR / extraction — the silent failure mode that breaks your RAG.
Rebuild reading order from positioned text boxes: strip running heads by cadence, rejoin paragraphs across page breaks, keep citations exact
To associate your repository with the reading-order topic, visit your repo's landing page and select "manage topics."