PulseAugur
EN
LIVE 09:10:58
中文(ZH) 一次吃下一本书!百度开源新OCR,作者疑似前DeepSeek研究员

Baidu releases Unlimited OCR, challenging long-context AI memory mechanisms · 1 source tracked

Baidu has open-sourced a new OCR model called Unlimited OCR, which excels at processing long documents by mimicking human reading habits. Unlike traditional OCR systems that process documents page by page and then stitch results together, Unlimited OCR uses a novel Reference Sliding Window Attention (R-SWA) mechanism. This allows it to maintain a continuous reading state without the memory and computational overhead that typically increases with document length, setting a new state-of-the-art on the OmniDocBench benchmark. AI

IMPACT Introduces a novel approach to long-context AI memory management, potentially impacting various sequence-based AI tasks beyond OCR.

RANK_REASON New OCR model release from a major tech company (Baidu) with novel attention mechanism and benchmark performance claims. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on 量子位 (QbitAI) →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Baidu releases Unlimited OCR, challenging long-context AI memory mechanisms · 1 source tracked

COVERAGE [1]

  1. 量子位 (QbitAI) TIER_1 中文(ZH) · 林樾 ·

    Eat a whole book in one go! Baidu open-sources new OCR, author suspected to be former DeepSeek researcher