Ocrs: Rust library and CLI tool for OCR (extracting text from images) (github.com)

Ocrs is a new Rust library and CLI tool for optical character recognition that aims to be a modern, ML-first alternative to engines like Tesseract. Announced as an early preview, it focuses on working well across diverse inputs—scans, photos, screenshots—while minimizing manual preprocessing by putting neural networks deeper into the pipeline. It’s designed for easy cross‑platform use (including WebAssembly), a clear and modifiable Rust codebase, and training on open, permissively licensed datasets, making it appealing for researchers and engineers who want an auditable, extensible OCR stack. Technically, ocrs uses PyTorch-trained neural models (also provided in ONNX for other runtimes) with a separate ocrs-models repo containing datasets and training tools. The CLI is installable via cargo install ocrs-cli and auto-downloads models to ~/.cache/ocrs on first run; it supports plain text output, JSON layout extraction, and annotated PNGs for word/line localization. The project includes unit and end‑to‑end tests (make check, make test-e2e) and can be run from source with cargo run. Current limitations: early preview quality (expect more errors than commercial OCR), Latin-alphabet support only, and a Rust toolchain requirement—future updates plan broader language coverage and continued model improvements.