昨天，DeepSeek 发布了一个新模型 DeepSeek-OCR。 这是一个专门为 OCR（文字识别）微调的 6.6GB 模型，主要贡献在于首次量化 “视觉 - 文本 token 压缩比”，验证 10× 近无损压缩、20× 仍保有 60% 精度的可行性；提出 DeepEncoder，解决现有编码器 “高分辨率 - 低内存 - 少 ...

Tesseract, the engine, was developed between 1985 and 1995 by HP Labs, but was tucked away when the company pulled out of the OCR business. Google called on the Information Science Research Institute ...