1.15.0 — Selective OCR for scanned and mixed PDFs

Packages

All packages were built from source checkpoint 06a9bab.

• Rust crate pdf-inspector 1.15.0 — cargo add pdf-inspector@1.15.0
• Python pdf-inspector 1.15.0 — pip install pdf-inspector==1.15.0
• Node @firecrawl/pdf-inspector 1.15.0 — npm install @firecrawl/pdf-inspector@1.15.0
• WebAssembly @firecrawl/pdf-inspector-wasm 1.15.0 — npm install @firecrawl/pdf-inspector-wasm@1.15.0

Node platform packages

• Linux x64 glibc
• Linux x64 musl
• Linux ARM64 glibc
• Linux ARM64 musl
• macOS ARM64
• Windows x64 MSVC

Highlights

• Adds selective OCR for scanned pages and mixed PDFs using PDFium rendering and PP-OCRv6 Small through ONNX Runtime (#348, #357, #358).
• Recovers credible PDFium text layers before OCR, routes only affected pages, and adaptively fuses OCR with usable native text (#359, #393, #394).
• Reuses OCR sessions and preserves row boundaries for faster, cleaner extraction (#390, #391).
• Exposes the OCR pipeline through Rust, pdf2md, Node.js, and Python, including page-level provenance, confidence, timings, warnings, and hosted-parser recommendations (#360, #403, #405).
• Adds pinned runtime and model caching, offline operation, and cross-language smoke coverage (#404, #409).

Other improvements

• Rejoins words hyphenated across line breaks (#388).
• Improves running-header, running-footer, and numbered-reference-note handling (#374, #392).

Compare changes since 1.14.2