Unlimited OCR 2 本地实测:单次解析整份文档且无页数限制
Baidu 发布 Unlimited OCR 2,可单次解析整份文档且不设页数限制,并通过 SR-SWA 仅聚焦关键 token。作者本地测试了手写物理公式、数学解答和发票表格,均可在数秒内解析;百页文档基准表现领先,但混合语言手写、花体旧印刷体和空表格行仍可能导致失败。
🚀 Baidu just dropped Unlimited OCR 2, and I tested it locally
📄 One model that parses whole documents in a single pass, with no page limit
🔹 New SR-SWA trick: it focuses only on the tokens that matter, so speed and memory stay nearly flat on long docs
🔹 Authors report big wins on hundred-page benchmarks, where rivals fall far behind
🔹 My hands-on tests: handwritten physics equations, math solutions and invoice tables parsed in seconds
🔹 Honest limits: mixed-language handwriting, ornate old print and empty table rows can trip it up
🎬 Watch the full video below 👇
piped.video/IAnxkqozuNs?si=vjfR…
@Baidu_Inc @Baidu_ERNIE
来源:fahdmirza · x.com