Native OCR plug-in
CraZY222123/dsh-ocr-plugin · model-adapters
A local OCR plug-in that adds "eyes" to DeepSeek Harness
★ 14GitHub stars
0Forks
2026-08-16Last updated
PythonLanguage
MITLicense
Key features
- Layout understanding (the layout principle transplanted from DeepSeek-OCR-2)
- Rearrange the reading order in row major order (left to right in multiple columns, top to bottom within columns)
- Grid-aligned table detection → revert to | separated Markdown lines
- Each line of text comes with the coordinates of the upper left corner, text@x,y, for the model to reorganize semantically.
Requirements
- This plug-in was developed and tested on Hongmeng PC (HarmonyOS PC), and has been partially adapted to HarmonyOS PC.
- RapidOCR is available normally in the Hongmeng PC sandbox environment
Install command
dsh plugin --profile web add github:CraZY222123/dsh-ocr-plugin