Baidu Proposes New Method for Long-Document OCR
burkov · x · 2026-07-12
This new paper from Baidu introduces Unlimited OCR, utilizing Reference Sliding Window Attention (R-SWA) to transcribe long documents in a single pass.
The core idea combines the highly compressed encoder of DeepSeek OCR with Baidu's own constant KV cache design, enabling the processing of dozens of pages in a single forward pass, even under the standard 32K context limit.
The authors also emphasize that R-SWA is not limited to OCR; it serves as a general-purpose parsing attention mechanism extensible to:
- ASR (Automatic Speech Recognition)
- Translation
- Other long-context parsing tasks
More from Research
- Deep Learning Course Adds Mechanistic Interpretability Lecture This Winter — CSProfKGD · 2026-10-26
- INCLUDE-2.0 benchmark evaluates regional knowledge across 91 languages and 86 countries — ABosselut · 2026-10-26
- CoRL paper: flow-matching VLAs can run 1-step Euler inference with no retraining — kaixhin · 2026-10-26
- CoVer at EMNLP 2026: Training Code LLMs to Generate and Self-Verify Simultaneously — algo_diver · 2026-10-26
- Robotic dexterity should be judged on speed, force control and tool use, not copying humans — DJiafei · 2026-10-26
- Turing-complete malware can run in PyTorch's backward pass via a zero-valued loss — retr0jirachi · 2026-10-26