OCR 識別OCR Recognition
研究性 demo，對清晰的印刷/寫本滿文頁效果最好。後端運行於單塊共享 GPU，已限流，請勿濫用。Research demo, best on clean printed/manuscript Manchu pages. Running on a single shared, rate-limited GPU — please don't abuse it.
OCR 模型目前只對楷書優化。如遇草書手稿、或背景複雜（例如帶水印）的頁面，模型能力可能大幅下降。若有大量類似頁面需要批量轉寫，歡迎致信給 corresponding author，可以專門協助處理、甚至按需優化模型對特定格式之優化。該模型為自學習範式，我們也在利用這一機制不斷提高模型對不同類型手稿的適應能力。The OCR model is currently tuned only for standard (kaishu) script; for cursive manuscripts or pages with complex backgrounds such as watermarks, accuracy can drop sharply. If you have many such pages to transcribe in bulk, feel free to write to the corresponding author — we can help process them, or even tune the model for your specific format. The model follows a self-learning paradigm, and we are continually leveraging this mechanism to improve its adaptability to different kinds of manuscripts.