DEEPOCR
// A reproduction of the Deepseek-OCR model including training
DeepOCR
A reproduction of the Deepseek-OCR model including training
13EmergingUnknown
What it does
A reproduction of the Deepseek-OCR model based on the VILA codebase. DeepOCR explores context optical compression through vision-text token compression, achieving competitive OCR performance with minimal vision tokens. ๐ Website | ๐ค Model - Token Efficiency: Achieves competitive OCR performance using ~250 vision tokens - Open Source Implementation: Complete reproduction of DeepSeek-OCR's
Getting Started
git
git clone https://github.com/pkulium/DeepOCR