OvisOCR: End-to-End Document Parsing via Aligning Specialized Perception with General Reasoning

This paper presents OvisOCR, a lightweight and strictly end-to-end Multimodal Large Language Model (MLLM) tailored for document parsing. Unlike current methods that rely on complex "Crop-OCR-Merge'" cascades to handle high-resolution inputs, OvisOCR directly maps full-page visual signals to structured Markdown without localized slicing or layout detection dependencies. Through extensive evaluations on the OmniDocBench, OvisOCR achieves top-tier performance among all compared methods in all aspects, demonstrating that a compact E2E model can effectively ``digest'' the capabilities of intricate pipelines and surpass specialized and general methods. Technically, OvisOCR unifies fine-grained recognition and semantic correction in a single model by leveraging supervision bootstrapped from strong OCR engines and refined via general-model-based data cleaning. To balance the performance across diverse document constituents, we design category-specific reward mechanisms for distinct element types, such as dense text, complex tables, and formulas, and ensure the model enhances its formatting strengths for each modality concurrently. This approach effectively resolves the optimization conflict, guaranteeing that improvements in structural layout parsing do not come at the expense of omitting fine-grained textual details. Empirical results confirm that OvisOCR eliminates the error propagation inherent in split-and-merge architectures, offering a streamlined path for future document intelligence. Our model is available at \url{https://huggingface.co/ATH-MaaS/OvisOCR}.

Many important documents—such as invoices, reports, forms, and academic papers—exist as images or PDFs, but computers need them converted into clean, structured text before they can search, edit, or analyze them. Existing document-reading systems often cut a page into small regions, read each region separately, and then stitch everything back together, which can introduce mistakes in reading order, tables, formulas, or dense text. We introduce OvisOCR, a lightweight AI model that reads a full document page directly and turns it into structured Markdown without relying on this cut-and-merge pipeline. To train it, we combine the strengths of specialized OCR systems, which are good at recognizing fine text details, with general AI models, which are better at checking whether the whole document is logically ordered and well structured. We also give the model separate learning signals for text, tables, and formulas, so improving one type of content does not harm another. Experiments show that OvisOCR achieves strong accuracy while remaining efficient. This matters because cleaner document parsing can make large collections of PDFs easier to search, reuse, and analyze. More broadly, it provides a simpler foundation for future document understanding systems.