@berryxia: Guys, I've done it! Fully open-sourced and free! I turned PP-OCRv6 directly into a local workstation, accelerated with CoreML on Mac, with one-click switching between Tiny, Small, and Medium model sizes! Tiny is only 1.5MB for extreme lightweight, Medium is 34.5MB for main...

Guys, I've done it! Fully open-sourced and free! I turned PP-OCRv6 directly into a local workstation, accelerated with CoreML on Mac, with one-click switching between Tiny, Small, and Medium model sizes! Tiny is only 1.5MB for extreme lightweight, Medium is 34.5MB focusing on accuracy, and Small strikes a balance in between. Supports image upload, batch processing, result export to CSV/Markdown/Excel, and automatic history saving. Everything runs completely locally, ensuring privacy and security—no data is uploaded. The best part is that it automatically enables CoreML acceleration on Apple Silicon, while Intel Mac and Linux can also run on CPU. I also made a browser-based Tiny model that works with zero dependencies, letting you use OCR directly in a web page. Comes with evaluation scripts to compare against OmniDocBench and macOS's built-in Apple Vision. Actual tests show good performance on tricky scenarios like curved surfaces, dot matrix fonts, and low contrast. Previously, the most annoying parts of local OCR were model downloads, environment setup, and the trade-off between accuracy and speed. Now I've packaged all of that up. Developers, researchers, and anyone who needs offline document processing can just clone and use it. This is actually something I built while dealing with many pitfalls in my daily OCR work. Hope it helps others with the same needs.