OCR models

[bropines]

Here I will write about new/existing OCR methods. Each post will be dedicated to a separate OCR module. If you want to request something that already exists, then write the following under this post:

1. Type of OCR. API or local launch (for example, Google LENS or Paddle OCR)
2. Link to the site\documentation\GitHub page
3. Your tests/metrics, if you did them.

Please note that some local models have already been rejected and will not be added. Here is the list:

1. EasyOCR - low quality compared to built-in models from MIT
2. Surya OCR (until a decent vertical OCR version for Japanese appears). Medium/high quality, but alas, there is a lot of garbage, and there is no normal software API for cleaning up Unlikely Text. (I'm a lazy ass)
3. Different types of local LLMs. Use OLLAMA

ToDo:

• - RapidOCR (Simplified PaddleOCR)
• - Fix BingOCR

[bropines]

New model: OneOCR

(From Microsoft, only works on Windows 10/11)

---

Quality:

• English, Russian, Japanese - tested, quality is above average
• The rest of the languages have not been tested (good luck)

Speed:
Very fast. Unfortunately, you cannot choose what to run it on. It is assumed that it will run on the CPU by default.

Parameters:

1. Newline_handling – you can remove or leave the line breaks.
2. Reverse_line_order - suitable for vertical Japanese, as vertical text is written from right to left.
3. No_UpperCase - needed for Latin text when the text is written in caps

Installation instructions

[bropines]

Detailed Installation Instructions:

1. Go to this website: https://store.rg-adguard.net
2. Enter the following into the search bar:

   https://apps.microsoft.com/detail/9wzdncrfjbh4

3. Download the file Microsoft.Windows.Photos_2025.XXXXX.XXXXX.X_neutral_~_8wekyb3d8bbwe.msixbundle (Replace the X's with the version numbers; I downloaded 11030.27002.0). It's crucial that the file is an msixbundle. Don't worry about the browser's warning about unsafe download; this is a workaround to download from the MS Store.
4. Open the downloaded msixbundle file using an archive manager (e.g., 7-Zip, WinRAR).
5. Inside the msixbundle, open the MSIX file (with your processor type, usually x64) using an archive manager as well.
6. Extract the following 3 files from the MSIX file:
   • oneocr.dll
   • oneocr.onemodel
   • onnxruntime.dll
7. Place these 3 files into the data/models/one-ocr folder.
8. Restart BT and select the new module.

Detailed Video Guide on Extracting Files:

msedge_CL9J6vJbGl.mp4